# Prompt Edit: Prompt Engineering Reference (full text) > Full text of every prompt framework guide, comparison and prompt flow on https://www.promptedit.app. Written by Sebastian Messingfeld. Index: https://www.promptedit.app/llms.txt ================================================================ # Prompt Frameworks (55) ================================================================ ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/aida [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › AIDA Framework Copywriting Beginner # AIDA Framework Use the world's most enduring copywriting formula — Attention, Interest, Desire, Action — to produce AI-generated marketing content that actually converts. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Attention, Interest, Desire, Action Origin Elias St. Elmo Lewis, 1898 Best for Marketing copy, ads, email, sales content Token cost Low to moderate **TL;DR:** AIDA is a four-stage copywriting formula — Attention, Interest, Desire, Action — that maps the psychological journey from noticing an ad to taking action. Use it for persuasive marketing content where conversion is the primary goal: ads, email campaigns, landing pages, and sales scripts. Label each stage explicitly in your prompt and supply concrete audience and product details for each section. ## What is the AIDA Framework? AIDA — Attention, Interest, Desire, Action — is the oldest surviving copywriting formula in marketing. First described by advertising pioneer Elias St. Elmo Lewis in 1898, it maps the four psychological states a prospect passes through on the way from ignoring you to buying from you. - **A — Attention:** Hook the reader immediately - **I — Interest:** Build engagement with relevance - **D — Desire:** Create emotional pull toward the outcome - **A — Action:** Drive a specific, friction-free next step Adapted for AI prompting, AIDA tells the model to structure persuasive content in a specific sequence: open with a hook that stops the reader mid-scroll (Attention), follow with information that earns continued reading (Interest), shift to emotional language that makes the reader want the outcome (Desire), and close with a specific, friction-free call to action (Action). What makes AIDA particularly powerful for AI use is that it encodes a proven psychological sequence into the prompt itself. Rather than asking the model to "write good marketing copy" — which leaves format and structure entirely to chance — AIDA gives the model a structural mandate that professionals use to achieve consistent conversion results. ## When to Use the AIDA Framework 📢 ### Digital Advertising Write Facebook, Google, or LinkedIn ads where you have seconds to earn a click and must move through the full persuasion sequence in very few words. 📧 ### Email Campaigns Structure promotional and nurture emails so the subject line hooks, the opening builds interest, the body creates desire, and the CTA captures action. 🌐 ### Landing Pages Build landing page copy that guides visitors through the buying psychology, from headline to social proof to offer to conversion button. 📱 ### Social Media Posts Craft posts that stop the scroll with an Attention hook and end with a specific action: comment, share, click, or save. 🎙️ ### Sales Scripts Structure sales call openers, voicemail scripts, and outreach messages that move prospects through interest and desire before making the ask. 📬 ### Cold Outreach Write B2B cold emails and LinkedIn messages that earn attention in a crowded inbox and convert enough interest into a reply or a meeting. ## How to Use the AIDA Framework 1. 1 ### Attention — Hook the reader immediately The first line must earn the second. Use a surprising statistic, a bold claim you can back up, a relatable frustration, or a counter-intuitive question. In your prompt, tell the model to open with a hook that speaks directly to the reader's situation. For ads: one sentence or headline. For emails: subject line plus first line. 2. 2 ### Interest — Build engagement with relevance Now that you have their attention, keep it by showing you understand their world. Feature benefits over features, demonstrate relevant expertise, or present evidence that shifts their thinking. Instruct the model to use clear, jargon-free language matched to your audience's vocabulary. 3. 3 ### Desire — Create emotional pull toward the outcome Move from "this is interesting" to "I want this." Paint a vivid picture of the transformation, use testimonials or concrete examples, and address the reader's emotional drivers — not just their logical ones. Tell the model to make the reader feel the benefit, not just understand it. 4. 4 ### Action — Drive a specific, friction-free next step Close with one clear call to action matched to the reader's readiness. Avoid multiple CTAs — they split attention and reduce conversion. Tell the model what the action is (click, reply, book, download) and whether to include urgency, scarcity, or a risk-reducer like a free trial or guarantee. ## Prompt Examples AIDA — SaaS Product Email ``` Write a promotional email using the AIDA framework for a time-tracking SaaS tool aimed at freelance consultants. ATTENTION: Open with a hook about the hidden cost of unbilled hours — use a specific, relatable statistic or scenario. INTEREST: Explain how the tool automatically tracks time across apps and drafts invoices, focusing on time saved and accuracy gained. DESIRE: Paint a picture of the relief of never chasing unpaid invoices or reconstructing a week's work from memory. Include one short testimonial-style sentence from a happy user. ACTION: End with a single CTA to start a 14-day free trial, no credit card required. Keep the total email under 220 words. Conversational tone. No jargon. ``` AIDA — LinkedIn Ad Copy ``` Write a LinkedIn single-image ad using the AIDA framework. Product: An online course teaching mid-career professionals how to transition into UX design in 12 weeks. ATTENTION: Lead with a bold, empathetic statement about feeling stuck in a career that no longer excites you. INTEREST: Briefly describe what the course covers and who it is for — people with no design background who want a career change without going back to university full-time. DESIRE: Use a transformation statement: who they are now vs. who they will be after completing the course. Keep it emotional and specific. ACTION: One line. Drive them to a free intro webinar this Thursday. Create urgency with limited spots. Total copy: under 150 words. Punchy, direct, no corporate language. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Proven over 125 years across virtually every medium and market | Can produce formulaic copy if not guided with specific detail about audience and product | | Naturally produces complete persuasive arcs — no sections get skipped | The linear sequence does not fit all formats — some channels need non-linear structures | | Beginner-friendly with four intuitive, universally understood labels | Desire and Interest sections can blur without clear guidance in the prompt | | Scales from a two-line ad to a multi-page sales letter | | ## AIDA Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Write {{ content_type }} for {{ product }} aimed at {{ audience }}. Attention: Open with {{ hook }}. Interest: Show why it matters to them: {{ relevance }} Desire: Make them want the outcome: {{ key_benefits }} Action: End with one clear call to action: {{ call_to_action }} ``` ## Prompt Flows That Use AIDA Framework AIDA Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Market Maker](https://www.promptedit.app/prompt-flow/the-market-maker) · Marketing Copy ## Frequently Asked Questions What is the AIDA framework? AIDA stands for Attention, Interest, Desire, Action. It is a century-old copywriting formula, first articulated by Elias St. Elmo Lewis in 1898, that maps the psychological journey a reader takes from first noticing an advertisement to making a purchase. Adapted for AI prompting, AIDA tells the model to produce content that hooks the reader (Attention), holds them (Interest), creates emotional pull (Desire), and drives a specific behaviour (Action). Why is AIDA still relevant after 125 years? Because the human decision-making sequence it describes has not changed. People still need to notice something before they can engage with it, still need a reason to keep reading, still respond to emotional resonance before logic, and still need a clear prompt to act. AIDA works because it mirrors psychology, not because it follows marketing fashion. When should I use AIDA for AI prompts? Use AIDA when the primary goal is persuasion or conversion: ad copy, email subject lines and bodies, landing page headlines, sales scripts, product launch posts, and fundraising appeals. If the content's job is to move someone from passive awareness to active decision, AIDA is the right framework. What is the difference between Interest and Desire in AIDA? Interest is rational engagement — you keep reading because the content is relevant and informative. Desire is emotional pull — you want the outcome the content promises. A feature list creates Interest. A vivid description of life after buying the product creates Desire. Both are needed because people buy on emotion and justify with logic. Can I use AIDA for long-form content? Yes. AIDA scales from a two-line social ad to a full sales page. For long-form, each section expands: Attention becomes a full opening section with a story or provocative statement, Interest covers multiple benefits and evidence points, Desire includes testimonials and transformation stories, and Action spells out multiple ways to engage. The proportions change, but the sequence stays the same. How do I write the Attention section without being clickbait? A strong Attention hook creates genuine curiosity or signals clear relevance to the reader's situation — it does not deceive. Effective openers include a surprising statistic, a relatable pain statement, a counter-intuitive claim you can back up, or a question the reader has already asked themselves. Clickbait promises something the content cannot deliver; a good Attention hook promises something it does. Should I always end AIDA with a hard sell? No. The Action step should match the stage of the buyer journey. For cold audiences, the action might be 'read this article' or 'book a free call'. For warm audiences it might be 'buy now'. The Action should be the most natural next step given how much trust has been built, not necessarily the biggest possible ask. ## Origin & Sources 1. Wikipedia (2026). [AIDA (marketing)](https://en.wikipedia.org/wiki/AIDA_(marketing)). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/ape [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › APE Framework Structured Framework Beginner # APE Framework Three purposeful components — Action, Purpose, Expectation — that anchor every prompt in clear intent and audience awareness, producing outputs that actually fit the reader. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Action — Purpose — Expectation Best for Audience-facing content and communications Complexity Low — three intuitive components Token cost Low **TL;DR:** APE (Action, Purpose, Expectation) is a three-component prompt framework that links what you want done to why it matters and what quality looks like. Use it for audience-aware tasks — marketing copy, instructional content, professional writing — where understanding the purpose sharpens the output. Write one sentence per component, making the audience and desired outcome explicit. ## What is the APE Framework? The APE framework is a three-component prompt structure that emphasizes audience awareness alongside task definition. Its components are **Action** (what the AI should do), **Purpose** (who the output is for and why it exists), and **Expectation** (what the output should look like in terms of format, length, and quality). - **A — Action:** Define what to do with a strong verb - **P — Purpose:** Describe the audience and intent - **E — Expectation:** Specify format, length, and quality What sets APE apart from other three-part frameworks is the Purpose component. Most minimal frameworks ask you to describe the task and the desired outcome. APE specifically asks you to articulate the audience and the intent — forcing you to think about the reader before you think about the format. This seemingly small shift produces noticeably more targeted outputs, especially for marketing copy, educational content, and customer-facing communications. APE is particularly well-suited to content teams, educators, and marketers who produce audience-facing material regularly. The framework is quick to compose, easy to remember, and scales to complex multi-audience scenarios simply by expanding the Purpose component. ## When to Use the APE Framework 📣 ### Marketing Copy Define the copy action, describe the target buyer in Purpose, and set expectations for format, tone, and call-to-action to get copy that speaks directly to your ideal customer. 🎓 ### Educational Content Specify what to explain (Action), describe the learner's background and concerns in Purpose, and set the format and vocabulary level in Expectation for genuinely accessible explanations. 💬 ### Customer Support Responses Ask for a support reply (Action), describe the customer's situation and emotional state in Purpose, and expect a specific tone, length, and resolution approach in Expectation. 📱 ### Social Media Posts Request the post type, define the platform's audience in Purpose, and specify character limits, hashtag requirements, and engagement style in Expectation for platform-native content. 📄 ### Onboarding Documentation Ask for user guides or tutorials (Action), describe the new user's knowledge level in Purpose, and set expectations around structure, examples, and length for documentation that actually helps users succeed. 📨 ### Email Campaigns Define the email type and message (Action), describe the subscriber segment and their relationship with the brand in Purpose, and specify subject line, length, and CTA in Expectation. ## How to Use the APE Framework 1. 1 ### Action — Define what to do with a strong verb Start with a clear, specific action verb: write, explain, rewrite, summarize, generate, translate, draft. Follow the verb with the specific content type and key message or topic. This is not where you describe context — just the task itself. Keep Action focused and direct. 2. 2 ### Purpose — Describe the audience and intent This is the component that makes APE distinctive. Describe who will read or use the output: their role, expertise level, emotional state, and relationship to the topic. Then explain why this output needs to exist — what it should accomplish. A well-written Purpose turns a generic output into something that resonates with a specific person in a specific situation. 3. 3 ### Expectation — Specify format, length, and quality Tell the AI exactly what a successful output looks like. Cover: the structure (numbered list, short paragraphs, table), the length (word count, number of items), the tone (formal, conversational, reassuring), and any hard constraints (no jargon, must include a CTA, avoid certain words). Specificity here directly determines output quality. ## Prompt Examples B2B Product Launch — LinkedIn Post ``` Action: Write a LinkedIn post announcing a new product launch. Purpose: This post is for a B2B SaaS company launching a project management tool. The audience is mid-level managers and team leads at companies with 50–500 employees who struggle with scattered workflows and missed deadlines. Expectation: A 150-word LinkedIn post with a compelling hook in the first line, three specific benefits of the product, a clear call to action linking to the product page, and a professional yet conversational tone. Include 3 relevant hashtags at the end. ``` ML Explanation for Non-Technical Users ``` Action: Explain the concept of machine learning to a complete beginner. Purpose: This explanation will be used in a FAQ section of a consumer app that uses ML for personalized recommendations. The reader has no technical background and may be slightly worried about AI making decisions about them. Expectation: A reassuring, jargon-free explanation of 120–150 words. Use one simple real-world analogy. End with a sentence explaining how the app uses ML to benefit the user specifically. Avoid the words "algorithm," "model," and "training data." ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Built-in audience awareness via the Purpose component | No Role component — add an expert persona manually if needed | | Simple enough to compose quickly for any content task | No dedicated example slot — works better for common content types | | Produces noticeably more targeted outputs than unstructured prompts | Less suited to technical, analytical, or reasoning-heavy tasks | | Scales naturally — richer Purpose sections yield richer outputs | | ## APE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Action: {{ action }} Purpose: {{ purpose_and_audience }} Expectation: {{ format_length_quality }} ``` ## Frequently Asked Questions What does APE stand for in prompt engineering? APE stands for Action, Purpose, and Expectation. Action defines what the AI should do, Purpose explains why it is being done and who it is for, and Expectation describes the desired output format and quality bar. The framework's strength is the Purpose component, which forces you to articulate the audience and intent behind a request. How does APE differ from TAG? Both are three-part beginner frameworks, but they emphasize different things. TAG (Task, Action, Goal) focuses on the task category, execution method, and outcome. APE collapses the task category into Action and replaces Goal with two more nuanced components: Purpose (audience and intent) and Expectation (format and quality). APE produces more audience-aware outputs; TAG produces more outcome-focused outputs. What should I put in the Purpose component of APE? Purpose should answer: who is this content or output for, and why does it need to exist? Include the target audience's characteristics (age, expertise level, emotional state, role), the context where the output will be used (a blog post, an internal report, a customer support email), and the goal it serves (educate, persuade, reassure, entertain). The more specific your Purpose, the more tailored the AI's response will be. Is APE better than RACE for content writing? APE and RACE serve overlapping but distinct needs. APE is stronger when audience awareness is the critical variable — marketing copy, educational content, customer communications — because Purpose makes you articulate exactly who you are writing for. RACE is stronger when you want to specify an expert persona and need explicit context about the situation. Many content writers combine both approaches by adding a Role line to an APE prompt. Can APE be used for technical tasks like code generation? Yes, though you will want to be specific in each component. Action should name the programming task precisely, Purpose should describe the codebase, team, or audience that will use the code, and Expectation should specify language, framework, code style, whether to include tests, and how much explanation to provide alongside the code. How detailed does the Expectation component need to be? As detailed as the task requires. For simple tasks (a short email, a tweet), one or two sentences about length and tone suffice. For complex outputs (a structured report, a multi-section article, a function with tests), spell out the format precisely: heading structure, word count per section, required elements, tone, and any hard exclusions. Under-specified Expectations are the most common cause of disappointing AI outputs. What kinds of tasks is APE best suited for? APE excels at audience-facing content tasks: marketing copy, educational explanations, customer communications, onboarding materials, social media posts, and any task where knowing the reader matters as much as knowing the task itself. It is less ideal for tasks without a clear audience (internal data analysis, code generation for personal projects) where RACE or RTF may be more natural. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/bab [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › BAB Framework Copywriting Beginner # BAB Framework Before describes the pain, After paints the goal, Bridge is the prompt — a copywriting formula that gives AI rich context and clear direction for any transformation task. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Before — After — Bridge Best for Marketing copy, coaching, and persuasion Complexity Low — narrative structure anyone can apply Token cost Low to Moderate **TL;DR:** BAB (Before, After, Bridge) structures persuasive content around a transformation — the painful current state, the desirable future state, and the Bridge showing how to get there. Use it for sales copy, pitch narratives, and change-communication pieces where emotional contrast drives action. Describe each state vividly and make the Bridge explicitly your product or solution. ## What is the BAB Framework? The BAB framework is a three-part prompt structure borrowed from direct response copywriting. Its components are **Before** (the current problem or pain state), **After** (the desired future state), and **Bridge** (the actual prompt asking for how to get from Before to After). The formula encodes context, goal, and task into a natural narrative arc that AI models respond to extremely well. - **B — Before:** Paint the current pain state - **A — After:** Define the desired future state - **B — Bridge:** Request the path between the two states The power of BAB lies in the Before-After pairing. Instead of abstractly describing context and goals, you paint two concrete pictures: where things are now and where they need to be. This gives the AI model a rich understanding of stakes, constraints, and success criteria without requiring a separate context section. The Bridge then focuses purely on the solution request, free from explanatory overhead. While BAB originated in marketing, it is remarkably versatile. It works for business strategy prompts, coaching and personal development tasks, product redesign briefs, conflict resolution strategies, and any scenario where you can articulate a clear gap between the current state and the desired state. If you can describe a transformation, BAB can help you prompt for it. ## When to Use the BAB Framework 📣 ### Marketing & Sales Copy Describe the customer's current frustration (Before), their aspiration (After), and ask for copy that bridges them — landing pages, email sequences, ad copy, and pitch decks. 🎯 ### Conversion Optimization Use Before to describe low-performing metrics and user behavior, After to state the target metrics, and Bridge to request a structured improvement plan or A/B testing strategy. 🧭 ### Career & Business Coaching Lay out the current professional situation (Before), describe the career or business goal (After), and ask for a step-by-step plan, script, or strategy in the Bridge. 🔄 ### Process Transformation Describe a broken or inefficient workflow (Before), articulate the target state (After), and request a process redesign, automation strategy, or change management plan as the Bridge. 💡 ### Product Messaging Before: customer pain without the product. After: customer success with it. Bridge: write the positioning statement, value prop, or elevator pitch that connects the two. 🤝 ### Negotiation & Conflict Resolution Describe the current conflict or disagreement (Before), the desired resolution (After), and ask for talking points, a communication framework, or a mediation approach as the Bridge. ## How to Use the BAB Framework 1. 1 ### Before — Paint the current pain state Describe the current situation honestly and specifically. Include the problem, its symptoms, any relevant metrics, and the emotional or business impact. The more specific your Before, the more targeted the AI's Bridge will be. Vague Befores ("things aren't working") produce generic Bridges. Specific Befores ("conversion rate is 8%, exit surveys show low urgency") produce actionable plans. 2. 2 ### After — Define the desired future state Describe what success looks like with enough specificity to make it measurable. Include target metrics, behavioral changes, or qualitative outcomes where possible. The After acts as the AI's north star — every element of the Bridge will be oriented toward achieving it. Be ambitious but realistic; stretch goals that feel unattainable produce impractical Bridges. 3. 3 ### Bridge — Request the path between the two states Now ask for the content, plan, or strategy that gets from Before to After. Be specific about the format you need (a 5-email sequence, a step-by-step framework, a script, a bulleted action plan), any constraints (tone, length, timeline), and any elements that must be included. The Bridge is where you exercise the most control over the output format. ## Prompt Examples SaaS Trial Conversion — Email Drip Sequence ``` Before: Our SaaS product's free trial conversion rate is 8%. Users sign up, explore the product for a few days, and then churn without upgrading. Exit surveys show they understand the features but don't feel urgency to pay. After: Free trial users convert at 25%+ because they clearly see the ROI of the product before the trial ends, feel a personalized connection to it, and experience a natural, low-pressure moment to upgrade. Bridge: Write a 5-email drip sequence for free trial users (days 1, 3, 5, 8, and 14). Each email should highlight one high-value feature with a real customer success story, include a specific CTA tied to that feature, and progressively build toward the upgrade ask. Tone: helpful, not pushy. Length: 150–200 words per email. ``` Freelancer Pricing Strategy ``` Before: I'm a freelance designer who undercharges clients. I quote projects based on gut feel, often come in too low, get scope creep, and end up earning below minimum wage when I account for revision rounds. I feel uncomfortable negotiating and usually cave when clients push back on price. After: I confidently quote projects using a systematic pricing method, build revision limits and scope boundaries into every contract, and hold my price during negotiations — resulting in 30–40% higher project revenue and far less stress. Bridge: Give me a step-by-step pricing framework I can use immediately. Include: a formula for calculating project rates based on time, expertise, and market positioning; exact language to use when presenting prices; a script for handling the "your rate is too high" objection; and a clause I can add to contracts to handle scope creep. Make it practical and direct — I need to use this within the week. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Encodes rich context naturally through the Before-After narrative | Less suited to purely technical or analytical tasks without a transformation narrative | | Exceptionally effective for persuasive and transformation-oriented tasks | Before and After sections can become verbose if not disciplined | | Easy to write — you describe your situation, not abstract prompt components | No dedicated Role or Format components — add them manually if needed | | Produces emotionally resonant, purpose-driven outputs | | ## BAB Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Write {{ content_type }} for {{ audience }}. Before: {{ current_pain }} After: {{ desired_future }} Bridge: Show how {{ product_or_solution }} gets them from Before to After. ``` ## Frequently Asked Questions What does BAB stand for in prompt engineering? BAB stands for Before, After, Bridge. Before describes the current problem or pain state — where things are now and what is wrong. After describes the desired future state — what success looks like. Bridge is the prompt for how to get from Before to After. Originally a copywriting formula, BAB is highly effective for persuasive content, transformation narratives, and problem-solving prompts. Where does the BAB framework come from? BAB originated as a copywriting formula used in direct response marketing and sales copy. Copywriters use Before-After-Bridge to create empathy (Before), paint an aspirational picture (After), and then present their product or solution as the Bridge. Prompt engineers adapted the structure because it naturally encodes the AI's context (Before) and goal (After) alongside the actual task (Bridge). What makes BAB different from other three-part frameworks like TAG or APE? TAG and APE describe a task and its requirements. BAB describes a transformation — a journey from a current state to a desired state. This makes BAB uniquely powerful for persuasive writing, coaching, and problem-solving tasks where the emotional arc matters as much as the information. The Before-After framing gives the AI rich context about stakes and motivation without you having to write a separate context section. Is BAB only for copywriting and marketing? No. While BAB has copywriting roots, it works excellently for any prompt involving a transformation: career coaching advice, business process improvement, personal development plans, product redesign briefs, and conflict resolution strategies. Whenever you can describe a clear gap between the current state and the desired state, BAB is an effective structure. How detailed should the Before and After components be? They should be specific enough to capture the real stakes and the concrete goal. Vague Befores ('things aren't working well') produce generic Bridges. Specific Befores ('conversion rate is 8%, users churn after 3 days, exit surveys show low urgency') produce actionable Bridges. The same applies to After — define measurable success criteria where possible so the AI knows exactly what solution to aim for. Can I use BAB for technical prompts like code or data analysis? Yes, though it is less natural than for narrative tasks. Before describes the current code or data problem, After describes the working system or insight you need, and Bridge asks for the implementation or analysis path. For purely technical tasks, RACE or RISEN may feel more efficient. BAB shines most when there is an emotional or motivational dimension to the transformation. How does BAB relate to the PAS (Problem, Agitate, Solution) copywriting formula? BAB and PAS are related copywriting formulas. PAS (Problem, Agitate, Solution) deliberately intensifies the problem before presenting the solution — it is more emotionally charged and works well for high-urgency sales copy. BAB is more neutral and balanced: it states the problem calmly, describes the ideal outcome positively, and requests the bridge. BAB tends to produce less hyperbolic AI outputs and is better suited to professional or coaching contexts. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/care [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CARE Framework Structured Prompting Beginner # CARE Framework Give the AI Context, an Action to perform, the desired Result, and a concrete Example — producing business communication that matches your voice and hits the right tone from the first attempt. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Context-Action-Result-Example Best for Cover letters, sales emails, persuasive writing Token cost Low **TL;DR:** CARE (Context, Action, Result, Example) is a four-component framework for professional writing that grounds every request in situational context and shows the AI what success looks like through an example. Use it for business communication, stakeholder emails, and persuasive documents where tone and style consistency matter most. Provide a short real-world example of the desired output style to anchor the AI's response. ## What is the CARE Framework? The CARE framework is a four-component prompt structure built for professional and persuasive writing. Its components — **Context**, **Action**, **Result**, and **Example** — guide the AI to produce communication that is grounded, purposeful, and stylistically consistent with your expectations. - **C — Set the Context** - **A — Define the Action** - **R — Describe the Result** - **E — Provide an Example** What sets CARE apart from similar frameworks is its explicit **Example** component. Rather than describing in words what good output looks like, you demonstrate it. A single well-chosen example paragraph gives the AI a style target that is far more precise than any abstract description. This makes CARE especially powerful for writing tasks where voice, tone, and register matter — cover letters, sales emails, board summaries, and client proposals. - **Context:** The situation, background, and relevant facts the AI needs to understand before writing. - **Action:** The specific writing task you want the AI to perform. - **Result:** The desired output format, length, tone, and quality criteria. - **Example:** A concrete sample of your writing style, a reference output, or an example that illustrates the target. ## When to Use the CARE Framework 📝 ### Cover Letters Provide your background as Context, the writing task as Action, the desired tone as Result, and a sample of your writing as Example to produce cover letters that genuinely sound like you. 📧 ### Sales and Outreach Emails Supply the prospect profile and pain point as Context, specify the email goal as Action, set length and CTA constraints in Result, and include a high-performing previous email as Example. 📊 ### Executive Summaries Provide full document content as Context, request a condensed summary as Action, define the format and executive audience as Result, and share a previous summary as Example to match the expected style. 🤝 ### Business Proposals Set the client situation and your solution as Context, request a proposal section as Action, specify the persuasive goal as Result, and provide a winning proposal excerpt as Example. 🌟 ### Recommendation Letters Describe the candidate and relationship as Context, request a specific type of recommendation as Action, define length and tone in Result, and include a strong reference letter as Example. 💬 ### LinkedIn and Social Content Provide your professional milestone as Context, request a post as Action, specify engagement-first formatting in Result, and share a high-performing previous post as Example. ## How to Use the CARE Framework 1. 1 ### Set the Context Provide all the background information the AI needs: who you are, who the audience is, the situation, key facts, and any constraints. The richer the context, the more relevant and accurate the output. Include quantified achievements, specific company names, and concrete details rather than vague descriptions. 2. 2 ### Define the Action State exactly what writing task you want performed. Be specific about the deliverable: "Write an opening paragraph" is better than "help me with my cover letter." A precise action prevents the AI from producing a generic template when you need something specific. 3. 3 ### Describe the Result Specify the output format, length, tone, and audience. Think of this as your acceptance criteria: what would a perfect output look like? Include must-haves (one clear call to action, under 150 words, no buzzwords) and must-avoids (passive voice, generic phrases). 4. 4 ### Provide an Example Share a concrete writing sample that demonstrates the style, voice, or quality you want. This can be your own previous writing, a piece you admire, or even a rough draft. Two to three sentences of authentic example material will improve output quality more than any amount of additional description. ## Prompt Examples Cover Letter Opening Paragraph ``` Context: I am a marketing manager with 8 years of experience in SaaS companies, applying for a Senior Marketing Director role at a Series B fintech startup. My most notable achievement is growing MQL volume by 340% in 18 months at my current company through account-based marketing. Action: Write a compelling opening paragraph for my cover letter that immediately communicates my value proposition and creates curiosity for the reader to continue. Result: A single paragraph of 3-4 sentences, confident but not arrogant, showing personality and clarity about why this specific role at this specific company matters to me. Example of my writing style: "Marketing without measurement is just noise. In my current role, I treat every campaign as a hypothesis — define the metric, run the test, double down on what works. That mindset turned a struggling ABM program into our top revenue channel." ``` B2B Cold Outreach Email ``` Context: Our B2B software company (project management tool) is reaching out to procurement managers at mid-market manufacturing companies (200-1000 employees) who are still using spreadsheets for production scheduling. Our tool reduces scheduling errors by an average of 67% based on customer data. Action: Write a cold outreach email for our sales team that leads with the prospect's pain point, presents our key proof point naturally, and closes with a low-friction call to action. Result: Email under 150 words, subject line included, conversational tone, no buzzwords, one clear call to action asking for a 20-minute call next week. Example of our brand voice: "Most scheduling problems are not scheduling problems — they are visibility problems. When your team cannot see bottlenecks forming in real time, small delays compound into missed deadlines. That is the problem we solve." ``` Resume Bullet Points ``` Context: I am a product designer applying to an in-house design lead role at a healthcare technology company. I have 6 years of experience designing mobile health apps and have led teams of up to 4 designers. The job description emphasizes patient-centered design and cross-functional collaboration with clinical teams. Action: Write three bullet points for my resume under my most recent role that highlight leadership, patient impact, and collaboration with non-designers. Result: Three bullet points starting with strong action verbs, each including a quantified outcome where possible, tailored to resonate with a healthcare-focused hiring manager. Example of resume language I like: "Championed accessibility-first design standards across 3 product teams, reducing support tickets related to usability by 41% over two quarters." ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Example component dramatically reduces style mismatches | Less effective for analytical or research-heavy tasks | | Ideal for professional and persuasive writing tasks | Requires you to have a good example ready — harder when starting from scratch | | Four intuitive components easy for anyone to use | Not designed for iterative workflows or multi-step reasoning | | Produces output that genuinely sounds like the writer, not a template | Output quality depends heavily on the quality of the Example provided | ## CARE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Context: {{ context }} Action: {{ action }} Result: {{ desired_result }} Example: {{ example_of_good_output }} ``` ## Frequently Asked Questions What does CARE stand for in prompt engineering? CARE stands for Context, Action, Result, and Example. Context sets the situation or background, Action specifies what the AI should do, Result describes the desired outcome or format, and Example provides a concrete sample to guide the AI's response. Together these four components produce persuasive, well-grounded business communication. When is the CARE framework most effective? CARE is most effective for persuasive and professional writing tasks: cover letters, business proposals, sales emails, executive summaries, case studies, and recommendation letters. Anytime you need the AI to write on your behalf in a context where first impressions and tone matter, CARE's Example component ensures the output matches your expected style. Why is the Example component important in CARE? The Example component is what differentiates CARE from simpler frameworks. By providing a concrete sample — a sentence in your voice, a paragraph in the right style, or an example output — you give the AI a quality target to match rather than an abstract description to interpret. This dramatically reduces mismatches between what you expect and what the AI produces. How does CARE differ from RACE? RACE stands for Role, Action, Context, Expectation. CARE replaces Role with a richer Context section and swaps Expectation for a concrete Example. CARE is better when you want to show the AI what success looks like through a sample; RACE is better when you want to define an expert persona and set quality criteria in words rather than by demonstration. Can I use CARE for emails and short-form content? Yes, CARE is particularly effective for emails, LinkedIn messages, cover letters, and other short-form professional content. The framework is lightweight enough that it does not pad the prompt unnecessarily, and the Example component ensures even short outputs hit the right tone and structure from the first attempt. What makes a good Example in the CARE framework? A good Example is specific and representative. It should show the AI the writing style, level of formality, sentence structure, and tone you want — not just the topic. For a cover letter, paste a paragraph of your previous writing you liked. For a sales email, share a previous email that achieved a high reply rate. The example does not need to be long — even two to three sentences can dramatically improve output quality. Is CARE suitable for technical documentation? CARE can be used for documentation, but it is not its strongest suit. The framework was designed for persuasive and communicative writing. For technical documentation, frameworks like RISE or RTF provide better structure for step-by-step instructional content. CARE works well for the human-facing, persuasive parts of documentation — such as introductions, product summaries, and executive overviews. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/chain-of-draft [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Chain of Draft Reasoning Intermediate # Chain of Draft (CoD) Reason step by step, but write each step as a terse draft of a few words. Chain of Draft keeps the accuracy benefits of Chain-of-Thought for a fraction of the tokens and latency. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Xu et al., Zoom Communications (2025) Builds on Chain-of-Thought Prompting Token savings Up to ~80% vs CoT (paper) Token cost Low **TL;DR:** Chain of Draft (Xu et al., 2025) is Chain-of-Thought with a word limit: the model reasons step by step but keeps each step to about five words, then gives the answer after a separator. In the paper it used roughly 80% fewer tokens than CoT on GSM8K at a small accuracy cost, and matched CoT on several other tasks. Use it with a few draft-style examples when cost and latency matter. ## What is Chain of Draft? Chain of Draft (CoD) is a prompting technique from the 2025 paper *Chain of Draft: Thinking Faster by Writing Less* by Silei Xu and colleagues at Zoom Communications. It starts from an observation about how people solve problems: we rarely write out every thought in full sentences. We jot down the essentials, such as a number, an equation or a keyword, and move on. Chain-of-Thought prompting makes models do the opposite. It produces long, verbose explanations, which cost tokens and time. Chain of Draft keeps the step-by-step structure but limits each step to a minimal draft, so the model still reasons explicitly while writing far less. The instruction used in the paper is short: ``` Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer at the end of the response after a separator ####. ``` On the GSM8K math benchmark, GPT-4o with Chain-of-Thought used about 205 tokens per answer at 95.4% accuracy. With Chain of Draft it used about 44 tokens at 91.1% accuracy, and latency fell from about 4.2 to 1.0 seconds. Claude 3.5 Sonnet showed a similar pattern. On date understanding and sports understanding tasks, Chain of Draft matched or exceeded Chain-of-Thought accuracy with a fraction of the tokens. ## When to Use Chain of Draft 💸 ### High-Volume API Workloads Classification, extraction and scoring pipelines where reasoning helps accuracy but verbose output multiplies your bill. ⚡ ### Latency-Sensitive Apps Chat assistants, autocomplete and voice interfaces where users notice every extra second of generated reasoning. 🧮 ### Arithmetic and Logic Word problems and calculations where the essential steps are numbers and short equations, not prose. 🎯 ### Cheap Self-Consistency Sampling several reasoning paths becomes affordable when each path is a few dozen tokens instead of a few hundred. 📱 ### Context-Limited Settings Long agent loops or small context windows where verbose reasoning would crowd out the information that matters. 🧾 ### Structured Decisions Rule checks, eligibility decisions and routing where each step is a condition and a result. ## How to Use Chain of Draft 1. 1 ### Add the draft instruction Tell the model to think step by step but keep each step to a minimal draft of about five words, and to put the final answer after a clear separator such as ####. 2. 2 ### Include two or three draft-style examples The paper found zero-shot CoD much weaker. Show examples whose reasoning is written as terse drafts (for example "20 - x = 12; x = 8") so the model copies the format. 3. 3 ### Parse the answer after the separator Because the reasoning is compact, extracting the answer is easy: split on the separator and keep what follows. 4. 4 ### Benchmark against Chain-of-Thought Run a small test set with both prompts. Keep CoD where accuracy holds and fall back to full CoT for the task types where it drops. ## Prompt Examples Few-Shot Chain of Draft for Math ``` Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer after ####. Q: Jason had 20 lollipops. He gave Denny some. Now he has 12. How many did he give Denny? A: 20 - x = 12; x = 8. #### 8 Q: A shop sells pens at 3 for $4. How much do 12 pens cost? A: 12 / 3 = 4 packs; 4 * 4 = 16. #### 16 Q: A train travels 180 km in 2.5 hours. What is its average speed in km/h? A: ``` Chain of Draft for a Business Rule Check ``` Decide whether this refund request is eligible. Think step by step, but keep each step to 5 words at most. Return ELIGIBLE or NOT ELIGIBLE after ####. Policy: Refunds within 30 days of delivery, item unused, not a final-sale item. Example Request: Delivered 12 days ago, unopened, regular item. Draft: 12 days < 30; unused yes; not final sale. #### ELIGIBLE Request: Delivered 41 days ago, unopened, regular item. Draft: ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Roughly 70–90% fewer reasoning tokens than Chain-of-Thought in the paper's benchmarks | Small accuracy drop on some tasks (about 4 points on GSM8K) | | Much lower latency, which users feel directly | Weak zero-shot: needs draft-style examples | | Keeps an explicit, checkable reasoning trace | Terse drafts are harder for humans to audit than full explanations | | Makes Self-Consistency and agent loops cheaper | Larger gap to CoT on small models (under ~3B parameters) | ## Chain of Draft Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer at the end after ####. Q: {{ example_question_1 }} A: {{ example_draft_1 }} #### {{ example_answer_1 }} Q: {{ example_question_2 }} A: {{ example_draft_2 }} #### {{ example_answer_2 }} Q: {{ question }} A: ``` ## Frequently Asked Questions What is Chain of Draft prompting? Chain of Draft (CoD) is a reasoning technique introduced by Xu et al. at Zoom Communications in 2025. Like Chain-of-Thought, it asks the model to reason step by step, but each step must be a minimal draft of about five words. The model keeps only the essential numbers, equations or facts and returns the final answer after a separator. The result is far shorter reasoning at similar accuracy on many tasks. What is the Chain of Draft prompt? The prompt used in the paper is: 'Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer at the end of the response after a separator ####.' The paper pairs it with a few worked examples written in the same terse draft style. How many tokens does Chain of Draft save? In the paper, Chain of Draft used as little as 7.6% of the tokens of Chain-of-Thought on some tasks. On the GSM8K math benchmark it cut reasoning tokens by roughly 80% (for example from about 205 to 44 tokens on GPT-4o) and latency by about 75%, while accuracy dropped from about 95% to about 91%. On other tasks, such as date and sports understanding, it matched or beat Chain-of-Thought. Is Chain of Draft better than Chain-of-Thought? It is cheaper and faster, not universally better. Use Chain of Draft when latency and token cost matter and a small accuracy trade-off is acceptable. Keep full Chain-of-Thought when the reasoning itself must be readable (for audits, teaching or debugging) or when every percentage point of accuracy counts. Does Chain of Draft work zero-shot? Not reliably. The authors report a significant drop in the zero-shot setting and conclude that the terse draft style is rare in training data, so the model needs a few examples written in draft style. Always include two or three short examples. Does Chain of Draft work with small models? Less well. The gap to Chain-of-Thought was more pronounced for models under about 3 billion parameters. Chain of Draft is best suited to capable frontier models. Should I use Chain of Draft with reasoning models? Reasoning models already think internally before answering, so an explicit visible draft adds little. Chain of Draft is most useful with standard instruction-tuned models where you pay for every visible reasoning token. ## Origin & Sources 1. Xu et al. (Zoom Communications) (2025). [Chain of Draft: Thinking Faster by Writing Less](https://arxiv.org/abs/2502.18600) . arXiv. 2. Wei et al. (2022). [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903) . arXiv / NeurIPS 2022. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/chain-of-thought [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Chain-of-Thought Prompting Reasoning Beginner # Chain-of-Thought Prompting (CoT) Guide AI models to reason through problems step by step — dramatically improving accuracy on math, logic, code, and any multi-step analysis. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as CoT, Step-by-step reasoning Origin Wei et al., Google Brain (2022) Best models Instruction-tuned GPT, Claude, Gemini (reasoning models do this natively) Token cost Higher (more output) **TL;DR:** Chain-of-Thought prompting instructs the model to show its intermediate reasoning steps before giving a final answer, dramatically improving accuracy on multi-step problems. Use it for math, logic, code debugging, and any task where an early reasoning error could corrupt the final result. Trigger it with "Let's think step by step" or provide a few worked examples that show explicit reasoning. ## What is Chain-of-Thought Prompting? Chain-of-Thought (CoT) prompting is a technique introduced by researchers at Google Brain in 2022. Instead of asking a model for a direct answer, CoT encourages the model to generate a sequence of intermediate reasoning steps — a "chain of thoughts" — that leads to the final response. The key insight is that explicitly verbalizing the reasoning process improves a model's ability to solve problems that require multiple logical steps. It mirrors how humans tackle hard problems: by writing down working notes rather than computing entirely in our heads. There are two main forms: - **Zero-Shot CoT:** Add "Let's think step by step" to your prompt — no examples needed. - **Few-Shot CoT:** Provide 2–8 examples that each include a reasoning chain before the answer. ## When to Use Chain-of-Thought Prompting 🧮 ### Math & Arithmetic Word problems, multi-step calculations, and equations where showing work reveals errors before they propagate. ⚖️ ### Logic & Deduction Syllogisms, cause-and-effect chains, legal analysis, and any problem requiring structured logical inference. 🐛 ### Code Debugging Trace execution paths, identify incorrect assumptions, and reason about state changes in a program step by step. 📊 ### Data Interpretation Analyze trends, compare figures, and draw conclusions from tables or charts with traceable logic. 🏥 ### Medical & Scientific Analysis Differential diagnosis, hypothesis testing, and interpreting research results where the reasoning chain matters as much as the conclusion. 🎮 ### Strategy & Planning Evaluate options, weigh trade-offs, and plan multi-step projects where decisions depend on earlier choices. ## How to Use Chain-of-Thought Prompting 1. 1 ### Identify if the task requires reasoning Ask yourself: "Would a human need to show their working to solve this?" If yes, use CoT. Simple factual questions don't benefit from step-by-step reasoning. 2. 2 ### Choose zero-shot or few-shot CoT For a quick win, add "Let's think step by step" to any prompt (zero-shot). For higher accuracy or a specific reasoning format, provide 2–5 worked examples with explicit reasoning chains (few-shot). 3. 3 ### Request the reasoning explicitly Tell the model to show its work. Phrases like "Explain your reasoning", "Walk me through each step", or "Think aloud" all trigger CoT behaviour. 4. 4 ### Extract and validate the final answer Ask the model to clearly state its conclusion after the reasoning chain: "After reasoning through this, state your final answer clearly." This separates reasoning from conclusion and makes verification easier. ## Prompt Examples Zero-Shot Chain-of-Thought ``` A store sells apples for $0.75 each and oranges for $1.20 each. Sarah buys 4 apples and 3 oranges. She pays with a $10 bill. How much change does she receive? Let's think step by step. ``` Few-Shot Chain-of-Thought ``` Example: Q: A train travels 60 mph for 2 hours, then 80 mph for 1.5 hours. What is the total distance? A: Step 1 — First leg: 60 mph × 2 h = 120 miles. Step 2 — Second leg: 80 mph × 1.5 h = 120 miles. Step 3 — Total: 120 + 120 = 240 miles. Answer: 240 miles. Now answer: Q: A project needs 12 developers. Each works 8 hours/day. The project requires 3,840 total developer-hours. How many working days will it take? ``` Code Debugging with CoT ``` The following Python function should return the factorial of n, but it returns wrong results for n > 1. Debug it step by step. def factorial(n): result = 0 for i in range(1, n): result *= i return result Walk through the execution for n=4, identify each bug, and provide the corrected code. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Significantly improves accuracy on complex tasks | Increases token usage and inference cost | | Makes model reasoning transparent and auditable | Reasoning chain can be wrong while answer looks correct | | Works zero-shot with just a trigger phrase | Limited benefit on smaller models | | Composable with role, few-shot, and self-consistency | Adds latency — not ideal for real-time applications | ## Chain-of-Thought Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` {{ task_or_question }} Think through this step by step before answering: 1. Restate what is being asked and list the relevant facts. 2. Work through the reasoning one step at a time, showing each intermediate result. 3. Check each step for errors before moving on. Finish with a line that starts with "Answer:" followed by the final answer only. ``` ## Prompt Flows That Use Chain-of-Thought Prompting Chain-of-Thought Prompting is one step in these 3 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Code Architect](https://www.promptedit.app/prompt-flow/the-code-architect) · Software Development - [The Error Detective](https://www.promptedit.app/prompt-flow/the-error-detective) · Debugging - [The Sprint Planner](https://www.promptedit.app/prompt-flow/the-sprint-planner) · Project Planning ## Frequently Asked Questions What is Chain-of-Thought prompting? Chain-of-Thought (CoT) prompting is a technique that encourages large language models to solve problems by generating a sequence of intermediate reasoning steps before giving a final answer. Instead of jumping straight to a conclusion, the model 'thinks out loud', which dramatically improves accuracy on complex tasks like math, logic, and multi-step analysis. When should I use Chain-of-Thought prompting? Use CoT when your task requires multi-step reasoning: math word problems, logical deduction, code debugging, legal or medical analysis, and any scenario where the path to the answer matters as much as the answer itself. For simple, factual queries, CoT adds unnecessary verbosity — stick to zero-shot for those. How do I activate Chain-of-Thought in a zero-shot setting? Simply append the phrase 'Let's think step by step' or 'Think through this carefully, one step at a time' at the end of your prompt. This zero-shot trigger reliably activates CoT reasoning in capable models like GPT, Claude, and Gemini without requiring any examples. What is the difference between zero-shot CoT and few-shot CoT? Zero-shot CoT uses a trigger phrase without providing any examples. Few-shot CoT provides 2–8 worked examples that each include both the reasoning chain and the correct answer, then poses the actual question. Few-shot CoT is generally more accurate but requires more careful example curation. Does Chain-of-Thought work on all AI models? CoT works best on large, instruction-tuned models with strong reasoning capabilities (current GPT, Claude and Gemini models). Smaller models may produce reasoning chains but still reach incorrect conclusions. The technique has diminishing returns below a certain model capability threshold. Dedicated reasoning models (such as OpenAI's o-series and GPT-5 thinking modes, Claude with extended thinking, or Gemini thinking models) already reason step by step internally; for them, an explicit 'think step by step' instruction adds little, and it is usually better to state the problem clearly and ask for a concise final answer. Can I combine Chain-of-Thought with other prompt styles? Absolutely. CoT combines excellently with Role Prompting ('As a senior accountant, let's think step by step...'), Few-Shot Prompting (providing reasoned examples), and Self-Consistency (generating multiple CoT paths and voting on the answer). These combinations often yield the best results on hard problems. What are the limitations of Chain-of-Thought prompting? CoT increases token usage and therefore cost and latency. The reasoning chain can also be incorrect while the final answer appears plausible — always validate CoT outputs on high-stakes tasks. Additionally, CoT may not help for tasks that are purely factual or creative without a logical structure. ## Origin & Sources 1. Wei et al. (2022). [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903) . arXiv / NeurIPS 2022. 2. Kojima et al. (2022). [Large Language Models are Zero-Shot Reasoners](https://arxiv.org/abs/2205.11916) . arXiv / NeurIPS 2022. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/chain-of-verification [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Chain-of-Verification Workflow Intermediate # Chain-of-Verification (CoVe) Make the model fact-check itself: draft an answer, turn its claims into verification questions, answer those questions independently, then rewrite the answer using only what held up. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Dhuliawala et al., Meta AI (2023) Goal Reduce hallucinations Steps Draft → plan checks → verify → revise Token cost High (several passes) **TL;DR:** Chain-of-Verification (Dhuliawala et al., Meta AI, 2023) reduces hallucinations in four steps: draft an answer, plan verification questions about its claims, answer each question independently of the draft, then write a final answer that keeps only verified claims. Use it for fact-heavy answers such as lists, biographies and research summaries. ## What is Chain-of-Verification? Chain-of-Verification (CoVe) was introduced in the paper *Chain-of-Verification Reduces Hallucination in Large Language Models* by Shehzaad Dhuliawala and colleagues at Meta AI. It targets one of the most persistent problems with language models: fluent, plausible statements that are simply wrong. The key insight is that models answer short, focused questions more accurately than they produce long answers full of facts. CoVe turns a long answer into a set of short questions, checks each one, and uses the results to repair the original answer: - **Draft:** generate a baseline response. - **Plan:** write verification questions that test each factual claim in the draft. - **Verify:** answer each question independently, without the draft in view, so the draft's errors are not simply repeated. - **Revise:** produce the final answer, correcting or dropping claims the verification did not support. The paper reports fewer hallucinations across list-based questions from Wikidata, closed-book question answering and long-form text generation. ## When to Use Chain-of-Verification 📋 ### Fact Lists "Name politicians born in New York" style questions, where a few wrong entries hide among correct ones. 👤 ### Biographies and Profiles Dates, roles and affiliations are exactly the details models tend to invent. 🔬 ### Research Summaries Check each claim about a paper, product or regulation before it reaches a report. 📰 ### Content With Numbers Articles and briefings where statistics, prices or dates must be right. 🧑‍💻 ### Technical Answers Verify API names, flags and version-specific behaviour before trusting generated instructions. 🏢 ### Customer-Facing Answers Support and sales answers where a confident wrong fact creates real cost. ## How to Use Chain-of-Verification 1. 1 ### Generate a baseline answer Ask the question normally. Treat the result as a draft, not a final answer. 2. 2 ### Plan verification questions Ask the model to list one short, specific question per factual claim in the draft (who, when, where, how many). 3. 3 ### Answer each question independently Ideally in separate calls without the draft in context. In a single prompt, instruct the model to answer from its own knowledge, not from the draft. 4. 4 ### Write the verified answer Compare the draft with the verification answers. Keep consistent claims, fix contradicted ones and remove anything that could not be confirmed. ## Prompt Examples Single-Prompt CoVe ``` Question: Which programming languages were created at Bell Labs, and in which year was each first released? Step 1 — Draft: Answer the question. Step 2 — Plan: For each language and year in your draft, write one verification question. Step 3 — Verify: Answer each verification question on its own, from your knowledge, without looking at the draft. Step 4 — Final: Write the final answer. Keep only claims the verification supports and mark anything uncertain. ``` Verification Step as a Separate Call ``` Answer each question independently and briefly. If you are not sure, say "uncertain" instead of guessing. 1. In which year was the C programming language first released? 2. Was C++ created at Bell Labs? 3. In which year was AWK first released? ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Measurably reduces hallucinated facts in the paper's benchmarks | Several passes: higher token cost and latency | | Makes each claim explicit and checkable | Relies on the model's own knowledge; cannot fix what it does not know | | Works with any model and needs no external tools | Independent verification needs separate calls for best results | | Pairs well with retrieval for high-stakes content | Less useful for opinion, creative or reasoning-only tasks | ## Chain-of-Verification Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Question: {{ question }} Step 1 — Draft: Answer the question. Step 2 — Plan: For each factual claim in the draft, write one short verification question. Step 3 — Verify: Answer each verification question independently, from your own knowledge, without relying on the draft. Say "uncertain" if you are not sure. Step 4 — Final: Rewrite the answer. Keep claims the verification supports, fix contradicted ones, and remove or flag anything uncertain. ``` ## Frequently Asked Questions What is Chain-of-Verification (CoVe)? Chain-of-Verification is a technique from Meta AI (Dhuliawala et al., 2023) for reducing hallucinations. The model drafts an answer, plans fact-checking questions about the claims in that draft, answers those questions independently, and then writes a final answer that keeps only what the verification supports. What are the four steps of Chain-of-Verification? 1) Generate a baseline response. 2) Plan verification questions that test the facts in that response. 3) Execute the verifications, answering each question independently so the answers are not biased by the draft. 4) Generate the final verified response, correcting or removing claims that failed verification. Why must verification questions be answered independently? If the model answers the verification questions while looking at its own draft, it tends to repeat the same mistakes. The paper found that the 'factored' variant, which answers each question separately without the original draft in context, removed more hallucinations than answering everything in one pass. Does Chain-of-Verification eliminate hallucinations? No. It reduces them, especially for lists of facts and long-form factual text, but it relies on the model's own knowledge. If the model does not know the correct fact, verification can confirm a wrong claim. For critical work, combine CoVe with retrieval from trusted sources or human review. How is CoVe different from Reflexion or the Prompt Refinement Loop? Reflexion critiques a failed attempt to learn a lesson for the next try, and the Prompt Refinement Loop improves the prompt itself. Chain-of-Verification targets factual accuracy within a single answer by turning its claims into questions and checking them. Can I use Chain-of-Verification in ChatGPT or Claude? Yes. The simplest version is one prompt with four labelled sections. For better results, run it as separate messages or API calls: draft, list questions, answer each question in a fresh context, then revise. ## Origin & Sources 1. Dhuliawala et al. (Meta AI) (2023). [Chain-of-Verification Reduces Hallucination in Large Language Models](https://arxiv.org/abs/2309.11495) . arXiv. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/clear [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CLEAR Framework Quality Framework Intermediate # CLEAR Framework Evaluate every prompt against five quality dimensions — Concise, Logical, Explicit, Adaptive, and Reflective — to produce professional, accurate, and self-verifying AI outputs for high-stakes work. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Concise-Logical-Explicit-Adaptive-Reflective Best for Technical docs, compliance writing, professional outputs Token cost Medium **TL;DR:** CLEAR (Concise, Logical, Explicit, Adaptive, Reflective) is a quality checklist for prompt writing, not a structural template — it defines how any prompt should be crafted rather than what sections to include. Use it for high-stakes work where accuracy and completeness are non-negotiable: legal, medical, financial, or compliance writing. Apply each principle as a final review pass before submitting your prompt. ## What is the CLEAR Framework? The CLEAR framework is a prompt quality checklist built on five principles that every well-crafted AI prompt should embody: **Concise**, **Logical**, **Explicit**, **Adaptive**, and **Reflective**. Unlike structural frameworks that define what components to include, CLEAR defines how any prompt should be written to maximize output quality and reliability. CLEAR is most valuable for professional and high-stakes work — technical documentation, compliance writing, financial analysis, medical information, legal summaries, and any context where accuracy, clarity, and completeness are non-negotiable. By running your prompt through the five CLEAR principles before submission, you systematically eliminate the most common causes of poor AI output: ambiguity, verbosity, missing context, and lack of self-verification. - **Concise:** Every word earns its place. No padding, no redundant context, no over-explained constraints. - **Logical:** The prompt follows a clear, coherent structure. Task, context, and constraints are organized so the model can parse them without ambiguity. - **Explicit:** Nothing is implied. All requirements, constraints, format expectations, and quality criteria are stated directly. - **Adaptive:** The prompt accounts for the conversation context, model capabilities, and any relevant prior output — building on what is already established. - **Reflective:** The prompt instructs the model to verify, self-critique, or flag uncertainty before presenting the final output. ## When to Use the CLEAR Framework 📋 ### Technical Documentation CLEAR's Explicit and Reflective principles ensure technical docs cover every required concept and verify accuracy before publication — critical when developers rely on the output to build systems. ⚖️ ### Compliance and Legal Writing The Explicit component prevents ambiguous legal language; the Reflective component builds in uncertainty flagging for areas where expert advice is needed rather than AI interpretation. 🏥 ### Medical and Scientific Content CLEAR's Logical and Reflective principles are essential in medical writing — ensuring claims are appropriately qualified, uncertainty is flagged, and the reader is directed to professional guidance for individual decisions. 💰 ### Financial Analysis and Reporting Concise and Explicit principles eliminate weasel words from financial reports; Reflective self-verification catches arithmetic errors and unsupported projections before they reach stakeholders. 🖊️ ### UX Writing and Interface Copy CLEAR's Concise and Logical principles are perfectly aligned with UX writing standards — every word must be justified, and the user's mental model must flow logically through the interface. 🎓 ### Academic and Research Writing Apply CLEAR's Logical structure to research arguments, use Explicit to define methodology and scope, and use Reflective to build in the self-critical review that peer-reviewed writing demands. ## How to Use the CLEAR Framework 1. 1 ### Write your prompt, then check for Concision Draft your prompt normally, then edit it ruthlessly. Remove every word that does not add information. Replace multi-word phrases with single precise words. Eliminate preamble like "I would like you to please" in favor of direct instructions. A concise prompt is typically 20-40% shorter than a first draft without losing any essential information. 2. 2 ### Verify Logical structure Read your prompt as if you are the AI. Does it flow from context to task to constraints in a logical order? Are there any contradictions between different parts of the prompt? Is the most important instruction easy to identify? Restructure if needed so the prompt has a clear hierarchy: who you are addressing, what you need, and what constraints apply. 3. 3 ### Make everything Explicit Scan for implied requirements. If you want the output in a specific format, say so. If certain words or claims are off-limits, list them. If the audience has specific knowledge assumptions, state them. If quality criteria apply (minimum examples, required sections, specific length), make them explicit. Anything left implicit is left to chance. 4. 4 ### Add Adaptive context and a Reflective verification step Check whether your prompt references relevant prior conversation context or model-specific capabilities. Then add a reflective instruction at the end: "Before responding, verify that all required sections are present and flag any claims where you have uncertainty." This single addition consistently improves output accuracy and completeness across all content types. ## Prompt Examples API Technical Documentation (all five CLEAR principles applied) ``` You are a senior technical writer with expertise in API documentation. Write the Getting Started section for a REST API that handles payment processing. The audience is backend developers who are already familiar with REST and HTTP but have not used this specific API before. The section must cover: authentication (OAuth 2.0, token-based), the base URL and versioning convention, the structure of a standard request and response, and error handling conventions. Use a code example for each concept. Format: markdown with H3 subheadings, code blocks for all examples, plain English explanations before each code block. Before submitting, verify that: every code example is syntactically correct, no OAuth flow steps are skipped, and the error handling section includes at least one 4xx and one 5xx example. ``` GDPR Compliance Summary with Uncertainty Flagging ``` You are a compliance analyst specializing in GDPR for SaaS companies. Summarize the key obligations a UK-based SaaS company faces when processing EU customer data under GDPR post-Brexit. Focus only on practical operational obligations — not legal history or political context. Be explicit: state whether each obligation applies to data processors, data controllers, or both. Flag any areas where the regulatory position remains uncertain or where expert legal advice is essential rather than this summary. Length: 400-500 words. Bullet points for obligations, plain language throughout. After completing the summary, review it and flag any statements where you have less than high confidence in accuracy, noting them clearly so the reader can seek further verification. ``` UX Copy Review with Explicit Criteria ``` You are a UX writer reviewing interface copy. The following button labels, error messages, and empty states are from a project management app. Evaluate each piece of copy against these three criteria: clarity (does the user immediately understand what will happen), action orientation (does it use active, directive language), and conciseness (is every word necessary). Copy to evaluate: - Button: "Click here to proceed with deletion of the selected items" - Error: "An error has been encountered" - Empty state: "It looks like there is nothing here yet. You can add items using the button above." For each item: state what is wrong, rewrite it, and explain what the rewrite improves. Keep feedback direct. Do not soften criticism with filler phrases like "great attempt" or "this is mostly good." ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Works as a quality layer on top of any structural framework | Requires deliberate review of each prompt before submission — adds time | | Reflective principle significantly reduces factual errors and omissions | Reflective instructions increase token usage and response length | | Explicit principle eliminates the most common source of prompt misinterpretation | Not a structural template — does not tell you what components to include | | Applicable across all content types and domains | Adaptive principle requires awareness of conversation context that beginners may overlook | ## CLEAR Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` {{ your_prompt }} Before answering, check this prompt against CLEAR: - Concise: remove anything that does not change the output. - Logical: the steps are in the order they should be done. - Explicit: format, length, audience and constraints are stated. - Adaptive: {{ context_that_may_change }} - Reflective: after answering, verify the output against {{ verification_criteria }} and fix any issues. ``` ## Prompt Flows That Use CLEAR Framework CLEAR Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Story Architect](https://www.promptedit.app/prompt-flow/the-story-architect) · Product Ownership ## Frequently Asked Questions What does CLEAR stand for in prompt engineering? CLEAR stands for Concise, Logical, Explicit, Adaptive, and Reflective. It is a quality framework for prompt engineering — rather than structuring the content of a prompt, CLEAR defines the five principles that every high-quality prompt should embody. Applying CLEAR means reviewing your prompt against each dimension before submitting it, ensuring it is tight, well-reasoned, unambiguous, context-aware, and self-correcting. How is CLEAR different from other prompt frameworks? Most prompt frameworks like RISE, RACE, or RTF are content templates — they tell you what components to include in a prompt. CLEAR is a quality checklist — it tells you how every prompt should be written, regardless of which structural framework you use. You can apply CLEAR principles to evaluate and improve a RACE prompt, a chain-of-thought prompt, or any other structured approach. What does Concise mean in the CLEAR framework? Concise means that every word in the prompt earns its place. Lengthy preambles, redundant context, and over-explained constraints dilute the signal and can confuse the model. A concise prompt says exactly what is needed — no more, no less. This does not mean short prompts are always better; it means that longer prompts should be long because the task genuinely requires more context, not because they are padded. What does Adaptive mean in the CLEAR framework? Adaptive means that the prompt is calibrated to the model's current context and capabilities. An adaptive prompt accounts for what the model already knows from the conversation, avoids re-explaining things already established, and adjusts complexity and detail to the specific model being used. In multi-turn conversations, adaptive prompting means building on previous responses rather than starting from scratch each time. What does Reflective mean in the CLEAR framework? Reflective means building self-correction mechanisms into the prompt. A reflective prompt asks the model to check its own work — to verify claims, identify gaps, flag uncertainty, and review the output before presenting it as final. This mirrors the human practice of reading back over written work. Reflective prompts consistently produce more accurate, complete, and honest outputs than prompts that request a single-pass answer. Can I use CLEAR with other prompt frameworks? Yes — CLEAR is specifically designed to complement rather than replace structural frameworks. After writing a RISE or TRACE prompt, run it through the CLEAR checklist: Is it Concise? Is the logic sound? Is the task Explicit enough? Is it Adaptive to the conversation context? Does it include a Reflective verification step? CLEAR is the quality layer on top of any structural framework. Is CLEAR useful for technical documentation and high-stakes writing? CLEAR is particularly valuable for high-stakes outputs — technical documentation, compliance writing, medical or legal content, financial analysis, and any domain where accuracy and clarity are non-negotiable. The Explicit and Reflective components are especially important here: being explicit about constraints prevents misinterpretation, and building in reflection catches errors before they appear in critical documents. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/coast [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › COAST Framework Structured Framework Intermediate # COAST Framework Build richly situational prompts by layering Context, Objective, Actions, Scenario, and Task — the framework that grounds AI outputs in realistic settings. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Context · Objective · Actions · Scenario · Task Signature feature Scenario component for situational realism Best for Training content, roleplay, case studies Complexity Medium — 5 structured components **TL;DR:** COAST (Context, Objective, Actions, Scenario, Task) adds a realistic Scenario component to standard structured prompting, grounding AI output in a specific real-world situation rather than abstract instructions. Use it for scenario-based planning, consulting deliverables, and training simulations where context realism shapes the quality of the response. Describe the Scenario with concrete details — industry, team size, constraints — before issuing the Task. ## What is the COAST Framework? COAST is a structured prompt engineering framework built around five components: **Context**, **Objective**, **Actions**, **Scenario**, and **Task**. Unlike purely instruction-driven frameworks, COAST includes a dedicated Scenario component that places the prompt inside a realistic, specific situation — making the AI's output far more contextually relevant. - **C — Context:** Set the Background - **O — Objective:** Define the Goal - **A — Actions:** List the Steps - **S — Scenario:** Ground it in a Situation - **T — Task:** Specify the Deliverable The framework is particularly powerful for content that needs to feel lived-in: training materials, customer service scripts, roleplay simulations, case studies, and any deliverable where the situational environment shapes the appropriate response. Think of COAST as the difference between asking "write a sales email" and providing the full situation — who the buyer is, what's at stake, what has already happened — so the AI can craft something that would actually work in that specific moment. ## When to Use the COAST Framework 🎭 ### Roleplay & Simulation Set the scene for negotiation practice, customer service training, interview prep, or conflict resolution scenarios with full situational context. 📚 ### Training Content Create realistic learning scenarios for onboarding, compliance training, or skill development where the situation determines the correct response. 📋 ### Case Studies Develop detailed case studies for business education, marketing, or internal reviews with full situational grounding and concrete deliverables. ✉️ ### Scenario-Based Writing Draft emails, proposals, reports, and communications that must account for a specific situation, relationship history, and contextual pressures. 🗺️ ### Customer Journey Mapping Map realistic customer scenarios across touchpoints, with specific personas, pain points, and goals that shape how communications should be written. 🏢 ### Organizational Planning Develop situationally-aware plans for hiring, restructuring, product launches, or crisis response where the specific organisational context drives the output. ## How to Use the COAST Framework 1. C ### Context — Set the Background Describe the environment, organisation, or situation the AI needs to understand before doing any work. Include size, industry, recent history, constraints, and any other background that shapes what good output looks like. This is the "who we are and where we are" section. 2. O ### Objective — Define the Goal State the overarching goal clearly in one or two sentences. What outcome are you trying to achieve? The Objective is the "why" behind the Task — it tells the AI what success looks like at a strategic level, even before the specific deliverable is named. 3. A ### Actions — List the Steps Break the work into numbered, concrete actions you want the AI to take. This is your instruction set — it removes ambiguity about scope and sequence. Actions should be specific enough that the AI knows exactly what to produce at each step. 4. S ### Scenario — Ground it in a Situation Describe the specific, realistic situation the output must address. Who are the people involved? What is their background, relationship, and current state? What pressures or constraints exist? The Scenario transforms generic output into something that could only work in this exact situation. 5. T ### Task — Specify the Deliverable State exactly what you want the AI to produce: a document, a list, a script, a plan. Be specific about format, length, and structure. The Task is your final instruction — everything above exists to inform this output. ## Prompt Examples Remote Onboarding Plan ``` Role: You are a corporate trainer designing onboarding materials. Context: Our company is a 200-person SaaS startup that recently shifted to fully remote work. New hires often feel isolated and struggle to understand internal processes during their first 30 days. Objective: Create a structured 30-day onboarding plan that helps new hires build relationships, understand company processes, and become productive quickly. Actions: 1. Outline week-by-week milestones for the first 30 days. 2. Include daily check-in rituals and async communication norms. 3. Specify who the new hire should meet in each week. 4. Add recommended tools, docs, and internal resources for each phase. Scenario: A new mid-level software engineer joins the team on Monday. They have strong technical skills but no prior remote-first experience. Their manager is in a different time zone. Task: Produce a day-by-day 30-day onboarding calendar with owners, resources, and success checkpoints for each week. ``` At-Risk Client Recovery ``` Role: You are a customer success manager at a B2B analytics platform. Context: Our platform helps e-commerce companies track conversion funnels. A long-standing enterprise client (3-year relationship, $180K ARR) has missed two check-in calls and their product usage dropped 40% over the last 60 days. Objective: Re-engage the client, identify the root cause of disengagement, and create a recovery plan that prevents churn. Actions: 1. Draft an outreach email that is warm but direct about the usage drop. 2. Prepare five discovery questions for the re-engagement call. 3. Propose three concrete value-add offers (e.g., feature workshop, custom report). 4. Define success metrics for the next 30 days post-call. Scenario: The client's primary champion recently left the company. The new point of contact is a VP of Marketing who is unfamiliar with your platform and skeptical of analytics tools. Task: Deliver the outreach email, the discovery question list, and a one-page recovery plan I can present internally before the call. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Scenario component produces situationally-aware, realistic output | Five components make it verbose — overkill for simple factual queries | | Actions section removes scope ambiguity and structures the work | Scenario writing requires effort to be specific without being overly constraining | | Works exceptionally well for training, roleplay, and case-study content | Less suited for purely analytical or data-processing tasks | | Flexible — components can be reordered or omitted for simpler tasks | | ## COAST Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Context: {{ context }} Objective: {{ objective }} Actions: {{ steps_to_take }} Scenario: {{ concrete_situation }} Task: {{ deliverable }} ``` ## Prompt Flows That Use COAST Framework COAST Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Story Architect](https://www.promptedit.app/prompt-flow/the-story-architect) · Product Ownership ## Frequently Asked Questions What does COAST stand for? COAST stands for Context, Objective, Actions, Scenario, and Task. Each component layers critical information: Context sets the background, Objective defines the goal, Actions list concrete steps, Scenario grounds the prompt in a realistic situation, and Task specifies the exact deliverable you need. How is COAST different from other structured prompt frameworks? The distinguishing feature of COAST is the Scenario component. While frameworks like RISEN or CRISPE focus on role and instructions, COAST explicitly grounds the prompt in a realistic, situational setting. This makes it particularly powerful for training content, roleplay, case studies, and any task where situational nuance matters. When should I use the COAST framework? Use COAST when your task requires situational realism — onboarding plans, customer service scripts, negotiation training, policy walkthroughs, or any scenario-based content. If your task is purely analytical or data-driven without a situational setting, a simpler framework like RISEN or APE may be more efficient. Can I skip components of COAST if they are not relevant? Yes. The Scenario component is the most frequently optional one — if your task is abstract rather than situational, you can omit it. However, the more components you include, the more precisely the AI can tailor its response. At minimum, Objective and Task should always be present. How detailed should the Scenario section be? The Scenario should include enough situational detail to remove ambiguity about the setting: who is involved, what their background is, what constraints or pressures exist. Two to five sentences is usually sufficient. Avoid over-specifying details that are not relevant to the output you need. Does COAST work well for creative writing and roleplay? Yes — COAST excels here. The Scenario component naturally sets the stage for roleplay, fiction, and immersive training simulations. You can use it to write customer service dialogues, negotiation scripts, interview prep scenarios, or even fictional short stories grounded in a specific setting. How does COAST compare to the RISEN framework? RISEN (Role, Instructions, Steps, End Goal, Narrowing) is more execution-focused and procedural. COAST adds a Scenario component that RISEN lacks, making COAST better for situationally-grounded content. RISEN is stronger for precise, multi-step technical instructions where situational context is less important. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/costar [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CO-STAR Framework Structured Prompting Intermediate # CO-STAR Framework The most popular framework for AI system prompt design — CO-STAR covers all six dimensions that determine how an AI writes: context, goal, style, tone, audience, and output format. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Context · Objective · Style · Tone · Audience · Response Best for Content creation, marketing, system prompts Popular with Claude & GPT system prompt designers **TL;DR:** CO-STAR (Context, Objective, Style, Tone, Audience, Response) is a six-component framework that controls not just what the AI says but how it says it — covering every dimension of voice and format. Use it for brand content, system prompt design, and audience-specific communication where style consistency is as important as accuracy. Specify each component explicitly, especially Style and Tone, to eliminate generic-sounding output. ## What is the CO-STAR Framework? CO-STAR is a six-component prompt engineering framework designed for content creation, communication tasks, and AI system prompt design. The acronym stands for **Context, Objective, Style, Tone, Audience, and Response** — covering every major dimension that determines not just what an AI says, but how it says it. - **C — Context:** Establish the operating situation - **O — Objective:** Define the communication goal - **S — Style:** Specify the writing structure and format approach - **T — Tone:** Set the emotional register - **A — Audience:** Characterize the reader precisely - **R — Response:** Define the output format and constraints **Where CO-STAR comes from:** the framework was developed by the Data Science & AI team at GovTech Singapore and became widely known after Sheila Teo used it to win Singapore's first GPT-4 prompt engineering competition in 2023, documenting the approach in [Towards Data Science](https://medium.com/data-science/how-i-won-singapores-gpt-4-prompt-engineering-competition-34c195a93d41). You will see it written as CO-STAR, COSTAR or CO STAR; all three refer to the same six components. The framework has become particularly popular in Claude and GPT system prompt design because it provides a complete specification language for AI assistant behavior. When you are configuring an AI assistant to work consistently across many different user inputs, CO-STAR gives you the vocabulary to define every relevant parameter: the operating context, the goal, the writing style, the emotional register, the target reader, and the output format. What distinguishes CO-STAR from frameworks like RISEN or QUEST is its explicit separation of **Style** and **Tone** as independent components. Most frameworks treat these as a single "voice" parameter. CO-STAR recognizes that the structural and technical aspects of writing (style) are distinct from the emotional register (tone) — and that getting both right simultaneously is what makes communication land with a specific audience. ## When to Use the CO-STAR Framework 🤖 ### AI System Prompts Configure Claude, GPT, or Gemini assistants with a complete behavioral specification — CO-STAR covers every dimension that shapes consistent, on-brand assistant behavior. 📣 ### Marketing & Brand Content Produce on-brand copy for a specific audience — social media posts, email campaigns, ad copy — where style, tone, and audience targeting are as important as the message itself. 📖 ### Long-Form Content Blog posts, whitepapers, and thought leadership articles where the style and audience calibration determine whether the content resonates or falls flat. 💬 ### Customer Communication Support responses, onboarding emails, and user-facing notifications where the tone must be precisely calibrated to the audience and context. 🎤 ### Speechwriting & Presentations Scripts, keynote talks, and executive presentations where the audience, tone, and style must be adapted to a specific speaker and occasion. 🌐 ### Multi-Channel Content Adapt a single piece of content across channels — LinkedIn, email, website — by changing the Response component while keeping Context, Objective, and Audience consistent. ## How to Use the CO-STAR Framework 1. 1 ### Context — Establish the operating situation Describe the situation the model is operating in: what product or service it represents, what the user relationship is, what has happened before this interaction, and any domain-specific knowledge the model should have. Context is the foundation everything else builds on. 2. 2 ### Objective — Define the communication goal State precisely what the output should achieve — inform, persuade, support, entertain, or convert. Go beyond the topic to the intended effect: "Help users solve problems quickly" is a better Objective than "Answer user questions about the product." 3. 3 ### Style — Specify the writing structure and format approach Define the structural and technical characteristics: formal or casual, dense or scannable, long-form or brief, technical or accessible. Include specific constraints like "use numbered steps for procedures" or "lead with data before explanation." 4. 4 ### Tone — Set the emotional register Specify the emotional quality of the writing: authoritative, empathetic, playful, urgent, warm, direct. Distinguish tone from style — you can be formal in style while being warm in tone. List specific words or phrases to avoid if tone calibration is critical. 5. 5 ### Audience — Characterize the reader precisely Define who will read the output in specific terms: role, industry, expertise level, what they care about, and what their biggest frustrations or goals are. The more specific your Audience definition, the more precisely the model can calibrate vocabulary, examples, and depth. 6. 6 ### Response — Define the output format and constraints Specify the format, length, structure, and any platform-specific constraints. Include document type (bullet list, numbered steps, essay), word count, heading structure, and any elements that must or must not appear in the output. ## Prompt Examples System Prompt — B2B SaaS Customer Support Assistant ``` Context: You are the AI assistant for Meridian, a B2B project management SaaS platform. Users are project managers and team leads at professional services firms. They use the platform daily and have intermediate to advanced familiarity with project management concepts. They typically contact support when something is blocking their work or when they cannot find a feature. Objective: Help users solve their problems quickly and completely. When a solution exists, provide it clearly and directly. When a limitation exists, acknowledge it honestly and offer the best available workaround. Style: Clear, direct, and practical. Use numbered steps for procedural answers. Use bullet points for options or comparisons. Avoid marketing language, filler phrases, and unnecessary caveats. Never say "Great question!" or "Certainly!". Tone: Professional and efficient, with warmth. Users are busy — respect their time. Be empathetic when they are frustrated, but do not dwell on apologies. Get to the solution quickly. Audience: Project managers and team leads at professional services firms. Assume familiarity with project management terminology but do not assume engineering knowledge. Use plain language for anything technical. Response: Keep responses under 250 words unless a step-by-step process genuinely requires more. Always end with a follow-up offer: "Let me know if you need help with anything else" or similar. Format code or settings paths in inline code format. ``` Content Creation — LinkedIn Thought Leadership Series ``` Context: This prompt is for a weekly LinkedIn post series about sustainable business practices for a B Corp certified management consulting firm. The series is called "The Responsible Business Brief" and runs every Tuesday. Previous posts have covered supply chain transparency, DEI measurement, and carbon accounting. Objective: Write a LinkedIn post for this week's topic — the business case for living wages. The post should inform, provoke thought, and position the firm as a credible, practitioner-led voice in the responsible business conversation. Style: Thought leadership essay style, compressed for social media. Lead with a counterintuitive observation or a data point. Use short paragraphs (2–3 sentences maximum). No bullet points. One clear call-to-action or question at the end to prompt comments. Tone: Confident and authoritative, but not preachy or activist. Speak from a business-outcomes perspective, not a moral-obligation perspective. The audience already cares about sustainability — the firm's job is to show them how to build the business case, not to convince them why they should care. Audience: CEOs, CFOs, and Chief People Officers at mid-to-large companies actively working on or interested in ESG and responsible business practices. Sophisticated, time-poor, skeptical of hype. They respond to data, case studies, and pragmatic frameworks. Response: 180–220 words. No hashtags in the body of the post — add 3–5 relevant hashtags as a separate line at the end. Include a specific data point or statistic to anchor the business case. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Separates Style and Tone as independent components — a precision other frameworks lack | Less suited for analytical or research tasks where communication style is not the primary variable | | Ideal for system prompt design — covers every dimension of persistent AI assistant behavior | Style and Tone components can overlap, requiring care to keep them distinct | | Audience component enables precise vocabulary and depth calibration | Six components add overhead for simple content requests where fewer parameters suffice | | Response component ensures output format is always explicitly specified | | ## CO-STAR Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Context: {{ context }} Objective: {{ objective }} Style: {{ style }} Tone: {{ tone }} Audience: {{ audience }} Response: {{ response_format }} ``` ## Prompt Flows That Use CO-STAR Framework CO-STAR Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Refiner](https://www.promptedit.app/prompt-flow/the-refiner) · Content Writing ## Frequently Asked Questions What does CO-STAR stand for in prompt engineering? CO-STAR stands for Context, Objective, Style, Tone, Audience, and Response. It is a six-component framework designed for content creation, communication, and audience-targeted writing. It is particularly popular for Claude and GPT system prompt design because it covers every major dimension of how a response should be written, not just what it should cover. Why is CO-STAR especially popular for system prompts? System prompts need to establish a persistent operating mode for an AI assistant across many different user inputs. CO-STAR covers every relevant dimension: what the assistant knows (Context), what it is trying to achieve (Objective), how it writes (Style and Tone), who it is writing for (Audience), and what its output should look like (Response). This comprehensiveness makes it the go-to framework for configuring AI assistants. What is the difference between Style and Tone in CO-STAR? Style is the structural and technical dimension of writing — formal vs. casual, technical vs. accessible, dense vs. scannable, long-form vs. brief. Tone is the emotional register — empathetic, authoritative, playful, urgent, reassuring. You might write in a formal style with a warm tone, or in a casual style with an urgent tone. Both dimensions are independent and together they define the full voice of the output. How specific should the Audience component be? The more specific your Audience definition, the more the model can calibrate its vocabulary, assumed knowledge level, and the examples it uses. 'Business professionals' is too vague. 'Finance Directors at mid-market B2B companies who are familiar with SaaS metrics but have no engineering background' allows the model to make precise choices about jargon, analogy selection, and depth of explanation. What goes in the Response component? Response defines the format and structure of the output: length, document type (bullet list, numbered steps, essay, table, JSON), section headings, and any formatting constraints. If you are building for a specific platform — a chat interface, a web page, a PDF report — the Response component is where you specify the constraints that make the output fit that context. Can CO-STAR be combined with role prompting? Yes — CO-STAR and Role Prompting are highly complementary. You can begin your system prompt with a role definition ('You are a senior product marketing manager...') and then use CO-STAR to specify the Context, Objective, Style, Tone, Audience, and Response for that role. The combination produces very consistent, well-characterized AI assistant behavior. Is CO-STAR better than RISEN or CRISPE for system prompts? CO-STAR's advantage over RISEN is its Style and Tone components, which make it particularly strong for content-heavy or communication-focused tasks. CRISPE adds a persona and voice component that overlaps with CO-STAR's Style. For system prompts where the communication style, audience targeting, and output format are the primary variables, CO-STAR is often the most precise fit. For task-oriented assistants, RISEN's Steps component offers stronger procedural control. ## Origin & Sources 1. Sheila Teo (2023). [How I Won Singapore's GPT-4 Prompt Engineering Competition](https://medium.com/data-science/how-i-won-singapores-gpt-4-prompt-engineering-competition-34c195a93d41) . Towards Data Science. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/create [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CREATE Framework Creative Framework Intermediate # CREATE Framework Guide AI through the full creative development lifecycle -- from initial concept and research through experimental variations to a polished, evaluated final output. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Conceptualize · Research · Experiment · Analyze · Transform · Evaluate Best for Innovation, creative development, iterative design Unique feature Experiment stage maximizes creative range before committing **TL;DR:** CREATE (Conceptualize, Research, Experiment, Analyze, Transform, Evaluate) is a six-stage framework that guides AI through the full arc of creative problem-solving from ideation to evaluated output. Use it for innovation challenges, product concepting, and complex creative briefs where a single-pass response is insufficient. Walk the AI through each stage sequentially, building on the previous stage's output before moving to the next. ## What is the CREATE Framework? CREATE is a six-stage prompt engineering framework that models the full arc of creative and innovation work. The acronym stands for **Conceptualize, Research, Experiment, Analyze, Transform, and Evaluate** — a sequential process that takes a creative challenge from initial ideation through systematic exploration to a polished, evaluated final output. - **C — Conceptualize -- Define the creative challenge and principles** - **R — Research -- Gather relevant context** - **E — Experiment -- Generate multiple distinct approaches** - **A — Analyze -- Evaluate and select the strongest approach** - **T — Transform -- Develop the winning approach fully** - **E — Evaluate -- Assess the final output against original goals** Most creative prompts ask for a single output on the first try. CREATE takes a different approach: it treats the AI as a creative collaborator that should explore multiple directions (Experiment) before narrowing to the strongest option (Analyze) and then developing it fully (Transform). This mirrors how skilled creative professionals actually work — breadth before depth, exploration before commitment. The framework is particularly powerful for innovation tasks, product design, campaign development, and any creative challenge where the correct answer is not obvious at the outset and exploring a range of possibilities before committing produces significantly better outcomes than generating a single response. ## When to Use the CREATE Framework 💡 ### Product and Feature Innovation Explore multiple product concepts or feature approaches before committing resources -- CREATE's Experiment stage surfaces options you might not have considered. 📣 ### Marketing Campaign Development Develop campaign concepts systematically -- from audience research through multiple concept experiments to a fully developed, evaluated campaign brief. 🎨 ### UX and Design Briefs Design onboarding flows, user journeys, or interface concepts by exploring multiple approaches before committing to a full specification. ✍️ ### Content Strategy Build content strategies from first principles -- research the competitive landscape, experiment with different editorial angles, and develop the strongest concept fully. 🏗️ ### Solution Architecture Explore multiple architectural approaches to a technical problem before analyzing trade-offs and developing the optimal solution in full detail. 🎓 ### Curriculum and Course Design Design learning experiences by conceptualizing learning goals, researching pedagogical approaches, and experimenting with different course structures before committing to a full curriculum. ## How to Use the CREATE Framework 1. 1 ### Conceptualize -- Define the creative challenge and principles State the goal, audience, constraints, and guiding principles for the creative work. This is the brief that all subsequent stages serve. Be specific about what success looks like and what values or principles should guide the creative process. 2. 2 ### Research -- Gather relevant context Direct the model to collect the context it needs before generating ideas: competitive landscape, established patterns, known failure modes, audience insights, or relevant case studies. Research grounds the creative work in reality and prevents the model from generating ideas that are already widely used. 3. 3 ### Experiment -- Generate multiple distinct approaches Ask the model to produce three to five genuinely distinct approaches, concepts, or variations -- not subtle variations of the same idea. Be explicit: "Make each concept genuinely different." This is the stage where creative range is maximized before narrowing begins. 4. 4 ### Analyze -- Evaluate and select the strongest approach Provide specific evaluation criteria and direct the model to rank the experimental options against them. Ask for a clear recommendation with rationale. This stage converts creative options into a defensible selection. 5. 5 ### Transform -- Develop the winning approach fully Direct the model to take the selected concept and develop it into a complete, polished output -- whether that is a full brief, a detailed specification, a complete draft, or a production-ready plan. Transform is where quantity of ideas becomes quality of execution. 6. 6 ### Evaluate -- Assess the final output against original goals Close the loop by asking the model to evaluate its own output against the goals set in the Conceptualize stage. Identify risks, gaps, or areas for improvement. This self-evaluation step often surfaces important refinements before you act on the output. ## Prompt Examples Marketing -- B2B Cybersecurity Content Strategy ``` Conceptualize: Develop a content marketing strategy concept for a B2B cybersecurity company targeting IT Directors at mid-market companies (200-1000 employees). Goal: establish thought leadership and generate 50 qualified leads per month within 6 months. Brand values: technical credibility, plain-spoken expertise, and a practitioner-first perspective. Research: Gather context on the current B2B cybersecurity content landscape -- what content formats are overused (avoid), what topics are underserved, and what channels IT Directors actually use to stay informed. Include insights on competitor content strategies if patterns are identifiable. Experiment: Generate three distinct content strategy concepts -- each with a different core content format, distribution channel emphasis, and editorial angle. Make each concept genuinely different, not slight variations of the same idea. Analyze: Evaluate all three concepts against the following criteria: differentiation from the current market, alignment with the brand's practitioner-first perspective, feasibility with a 2-person marketing team, and expected lead generation potential. Identify the strongest concept and explain why. Transform: Develop the strongest concept into a complete 6-month content strategy including: editorial calendar framework, 5 pillar content topics with 3 article ideas each, recommended distribution channels with posting cadence, and one flagship lead magnet concept. Evaluate: Assess the final strategy against the original goal -- does it credibly support generating 50 qualified leads per month within 6 months? Identify the biggest risk and suggest one mitigation approach. ``` Product Design -- Mobile Finance App Onboarding Experience ``` Conceptualize: Design a mobile app onboarding experience for a personal finance app targeting users aged 22-35 who have never used a budgeting tool before. Core objective: get users to their first successful budget setup within 10 minutes without feeling overwhelmed or judged about their current financial situation. Design principles: empathetic, simple, progress-focused. Research: Review current best practices in mobile onboarding for finance apps, including documented patterns from Monzo, YNAB, and Mint. Identify the most common failure points in fintech onboarding (drop-off moments) and what design solutions have been validated. Experiment: Propose three distinct onboarding flow concepts -- each with a different approach to the first-run experience, ranging from a minimal data-entry approach to a more guided, educational flow. Sketch the key screens or steps for each concept. Analyze: Evaluate the three concepts against: time-to-first-value (target: under 10 minutes), perceived complexity, emotional tone alignment with the empathetic design principle, and technical feasibility for a 3-person mobile dev team. Identify the optimal concept. Transform: Develop the optimal concept into a complete onboarding flow specification: screen-by-screen description, microcopy for key moments (welcome screen, permission requests, first budget setup), and a rationale for each design decision. Evaluate: Review the final onboarding spec against the original objective. Identify any step where user drop-off risk is highest and propose a specific design intervention to reduce it. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Experiment stage surfaces creative options that a single-output prompt would never reveal | Six stages generate significantly more tokens than simpler frameworks -- higher cost and latency | | Analyze stage creates a defensible, criteria-based selection rather than an arbitrary choice | Overkill for tasks with a clear brief where iteration is not needed | | Full six-stage arc ensures outputs are grounded, explored, developed, and validated | Requires careful criteria definition in the Analyze stage or the selection becomes arbitrary | | Adaptable to non-creative tasks -- any discovery-to-delivery workflow benefits from the structure | | ## CREATE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Conceptualize: {{ creative_challenge_and_principles }} Research: {{ relevant_context }} Experiment: Generate {{ number_of_approaches }} distinct approaches. Analyze: Evaluate each approach against {{ criteria }} and pick the strongest. Transform: Develop the winning approach fully. Evaluate: Critique the result and list what to improve next. ``` ## Frequently Asked Questions What does CREATE stand for in prompt engineering? CREATE stands for Conceptualize, Research, Experiment, Analyze, Transform, and Evaluate. It is a six-stage creative process framework that guides an AI model through the full lifecycle of a creative or innovation task — from initial ideation through to a polished, evaluated output. How is CREATE different from IDEA for creative tasks? IDEA is a four-component framework optimized for iterative refinement — it is best when you have a clear intent and want to converge on a polished output over multiple revision cycles. CREATE is a six-stage process framework that covers the full arc of creative development, from conceptualization through research and experimentation to a final evaluated output. Use IDEA when you have a clear brief; use CREATE when you are exploring from first principles. What happens in the 'Experiment' stage? Experiment directs the model to generate multiple distinct approaches, variations, or interpretations before committing to one. Instead of producing a single answer, the model produces a range of creative options — different angles, formats, or framings. This is where creative range is maximized before the Analysis stage narrows the field to the strongest options. Can CREATE be used for non-creative tasks? CREATE's methodology applies to any task with a discovery-to-delivery arc: product design, solution architecture, curriculum development, or business model innovation. The 'creative' in CREATE refers to the process of generating something new, not exclusively to artistic output. Any task that benefits from exploring multiple approaches before committing to one is a candidate for CREATE. What is the 'Transform' stage responsible for? Transform is where the best experimental approach from the Analysis stage is developed into a final, polished output. This might mean expanding an outline into a full document, refining a rough concept into a production-ready design brief, or taking a promising code sketch and building it into a complete, tested implementation. Transform is where quantity converts to quality. How do I write an effective 'Evaluate' directive in CREATE? An effective Evaluate directive specifies the criteria against which the final output should be assessed. It should mirror the original goals from the Conceptualize stage. For example: 'Evaluate whether the final concept is original enough to stand out in the current market, practical enough to execute within a 6-month timeline, and aligned with the brand values defined in the Conceptualize stage.' Is CREATE suitable for generating marketing campaigns? Yes — CREATE is especially well-suited for marketing campaign development. Conceptualize defines the campaign goal and audience; Research gathers competitive and cultural context; Experiment generates multiple campaign concepts; Analyze ranks them by strength; Transform develops the best concept into a full brief; Evaluate checks it against the brief's objectives. The result is a campaign concept that has been systematically developed rather than arbitrarily generated. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/crisp [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CRISP Framework Structured Prompting Beginner # CRISP Framework Five focused components — Context, Role, Instructions, Specifics, Persona — that transform vague task requests into precisely characterised, voice-consistent professional outputs. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Context · Role · Instructions · Specifics · Persona Best for Customer service, professional communication, task delegation Complexity Low — five intuitive components Output type Single, precise, voice-consistent response **TL;DR:** CRISP (Context, Role, Instructions, Specifics, Persona) is a five-component framework that gives the AI both expertise and a distinct voice, making it ideal for customer-facing and brand-critical content. Use it when consistent persona and professional tone matter as much as task accuracy — chatbots, support scripts, branded copy. Define Persona with specific communication traits like "empathetic but direct" to differentiate it from Role. ## What is the CRISP Framework? CRISP is a five-component prompt engineering framework designed for professional communication, customer-facing tasks, and any situation where voice and persona matter as much as task completion. The acronym stands for **Context, Role, Instructions, Specifics, and Persona** — five directives that together ensure the AI knows not just what to do, but who it is when it does it. - **C — Context:** Set the situational background - **R — Role:** Define the professional expertise - **I — Instructions:** State the task directly and completely - **S — Specifics:** Add the concrete constraints - **P — Persona:** Give the voice a human identity What distinguishes CRISP from similar frameworks like RACE or RTF is its **Persona** component. Where Role defines the professional position and domain expertise the AI draws on, Persona adds the human layer: the specific name, voice, communication style, and individual perspective the AI should write from. This distinction produces outputs that feel written by a specific person rather than generated by a generic professional, which matters significantly for customer communication, brand-voice content, and leadership communication tasks. CRISP is a streamlined version of the six-component CRISPE framework, with the Experiment stage removed. This makes it faster and more direct — ideal for tasks where you need a single precise output rather than a set of options to evaluate. Think of CRISP as the professional communicator's prompt: everything needed for a polished, characterised, situationally appropriate response, nothing extraneous. ## When to Use the CRISP Framework 📞 ### Customer Service Write responses to complaints, inquiries, and escalations that are situationally aware, empathetically toned, and branded — with the Persona component ensuring they read as coming from a real person, not a bot. 📧 ### Professional Emails Draft business emails — proposals, follow-ups, difficult conversations — from a specific professional persona with a defined voice and authority level, producing outputs that sound genuinely written rather than templated. 🤝 ### Task Delegation Define a task so precisely that the output requires no revision — the combination of Role, Instructions, Specifics, and Persona eliminates the most common sources of misaligned AI outputs. 🤖 ### AI System Prompts Configure AI assistants with a complete behavioral specification — CRISP maps naturally onto the five key dimensions of consistent assistant behavior: environment, expertise, behaviors, constraints, and voice. 🏢 ### Executive Communications Ghost-write internal memos, all-hands messages, or stakeholder updates from a specific leader's persona — with CRISP ensuring the output reflects their specific communication style and authority level. 📣 ### Brand Voice Content Produce on-brand marketing copy, social media posts, or web content from a precisely defined brand persona — ensuring every output reflects the same voice, regardless of which team member prompted it. ## How to Use the CRISP Framework 1. 1 ### Context — Set the situational background Describe the specific situation the AI is operating in: what has happened, what the current state is, what relationships or history are relevant, and any emotional or operational dynamics that should shape the response. Context is what makes an AI response feel situationally aware rather than generic. The more specific your context, the more precisely calibrated the output will be. 2. 2 ### Role — Define the professional expertise Tell the AI what professional position and domain expertise it should draw on. Include seniority level, specialisation, and any domain-specific knowledge that should inform the response. Role sets the knowledge base and problem-solving framework — it answers the question: what kind of professional would handle this situation, and what do they know? 3. 3 ### Instructions — State the task directly and completely Give a clear, active-verb directive for what the AI should produce. Be specific about what the output should accomplish, what it should include, and what the response needs to achieve for the reader. Instructions is where you state the job — make it unambiguous enough that two different people reading it would produce the same type of output. 4. 4 ### Specifics — Add the concrete constraints Layer in all the specific parameters that make the output precisely usable: word count limits, required inclusions (names, specific offers, phrases to use), tone prohibitions (no passive voice, no jargon), format requirements, and any constraints that turn a good answer into exactly the right answer. Specifics is where vague gets replaced with precise. 5. 5 ### Persona — Give the voice a human identity Name the specific person or character the AI is writing as, and describe their individual communication style, personality, and perspective. Persona goes beyond Role — it is the difference between "a senior customer service representative" and "James, who speaks directly, genuinely cares about resolution, and has authority to make decisions on the spot." Persona is what makes outputs feel authored, not generated. ## Prompt Examples Customer Service — Handling a Late Delivery Complaint ``` Context: A customer has just submitted a complaint saying their order arrived three days late and one item was missing. They have already sent two previous emails with no response and are clearly frustrated. Role: You are an empathetic senior customer service representative for a premium e-commerce brand. Your brand values: transparency, speed, and genuine accountability — never deflection. Instructions: Write a reply that acknowledges the delay and missing item without making excuses, apologises sincerely, explains the next steps clearly, and offers a meaningful resolution (full refund for the missing item plus a 20% discount on their next order as a goodwill gesture). Specifics: Keep the email under 200 words. Use the customer's name (Sarah) in the opening. Do not use passive voice or corporate phrases like "rest assured" or "we apologise for any inconvenience." Be direct and human. Persona: Write as James, a senior customer success manager who genuinely cares about getting this right and has the authority to make decisions on the spot. ``` Product Marketing — Landing Page Copy for a Developer Tool ``` Context: We are launching a new project management tool targeting remote-first software development teams at companies of 20-100 people. The market is crowded with tools like Linear, Jira, and Notion. Our differentiator is automated sprint retrospectives powered by AI — a feature no direct competitor offers. Role: You are a senior product marketer with deep experience in developer tools and SaaS go-to-market strategy. You have launched developer-focused products before and understand that this audience is highly allergic to marketing hype. Instructions: Write the hero headline and sub-headline for our product landing page. Then write a 60-word opening paragraph that communicates the core value proposition. Follow with three feature bullets (one sentence each) that speak to the product's most compelling capabilities. Specifics: The headline must be under 10 words. Avoid the words "powerful," "seamless," "next-generation," and "revolutionary." The copy should speak to the outcome (better team performance, less meeting overhead) not the mechanism (AI-powered retrospectives). Tone: direct, developer-respecting, confident without being brash. Persona: Write as a practitioner who has personally suffered through bad sprint retrospectives and knows exactly what teams actually need — not as a marketer describing a product from the outside. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Persona component produces outputs that feel written by a person, not generated by a system | No Experiment stage — add CRISPE's E component when exploring multiple options is valuable | | Five components are easy to memorise and apply quickly for everyday professional tasks | Role and Persona can blur without deliberate effort to keep them distinct | | Specifics component eliminates the most common sources of misaligned AI outputs | Less suited for analytical, research, or multi-step reasoning tasks — use RACE or RISEN instead | | Maps directly onto AI system prompt design for consistent assistant configuration | | ## CRISP Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Context: {{ context }} Role: You are {{ role }}. Instructions: {{ instructions }} Specifics: {{ constraints }} Persona: Write as {{ persona_voice }}. ``` ## Prompt Flows That Use CRISP Framework CRISP Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The API Designer](https://www.promptedit.app/prompt-flow/the-api-designer) · Software Development ## Frequently Asked Questions What does CRISP stand for in prompt engineering? CRISP stands for Context, Role, Instructions, Specifics, and Persona. It is a five-component structured prompt framework designed for professional communication, customer service, and task delegation tasks. Each component addresses a different layer of the prompt: the situational background, the AI's professional identity, the task directive, the concrete constraints, and the voice and perspective the AI should write from. How is CRISP different from CRISPE? CRISP is CRISPE with the Experiment (E) component removed. CRISPE asks the AI to brainstorm multiple options before settling on an answer, which is useful for creative or exploratory tasks. CRISP is designed for tasks where you need a single, precise, well-characterised output rather than a range of options — making it faster and more direct for customer service, professional communication, and delegation tasks. What is the difference between Role and Persona in CRISP? Role defines the professional position and domain expertise the AI should draw on — the knowledge base and problem-solving framework it should use. Persona defines the specific voice, perspective, and personal characteristics it should write from — the individual behind the role. A Role might be 'senior customer service representative'; the Persona would add 'named James, speaks directly, has authority to make decisions on the spot, genuinely cares about resolution.' What goes in the Specifics component? Specifics is where you add all the concrete constraints that make the output precisely usable rather than generically acceptable: word count, required inclusions (the customer's name, a specific offer amount, a particular phrase), format requirements (bullet points, numbered steps, plain prose), tone prohibitions (no passive voice, no corporate jargon), and any other parameters that turn a good answer into the right answer. When should I use CRISP over RACE or RTF? Use CRISP when character and voice matter as much as task completion — particularly for customer-facing communication, professional writing with a specific named persona, or tasks where the Persona component would meaningfully change how the output reads. RACE and RTF are better for analytical, research, or general task completion prompts where the AI's specific voice and persona are less critical to the output quality. Can CRISP be used for system prompts? Yes — CRISP maps very naturally onto AI system prompt design. Context sets the operating environment, Role defines the assistant's expertise, Instructions capture its primary behaviors, Specifics encode its formatting and constraint rules, and Persona establishes its voice and personality. The five components together produce a complete, consistent behavioral specification for an AI assistant. Is CRISP suitable for non-English prompts? Yes. CRISP works in any language. Complete all five components in the target language, and ensure the Persona component references cultural communication norms relevant to your context — directness levels, formality registers, and professional conventions vary significantly across languages and cultures. A Persona defined for a German business email will behave very differently from one defined for a Japanese customer service context. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/crispe-framework [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CRISPE Framework Structured Framework Intermediate # CRISPE Framework Six components — Capacity, Role, Insight, Statement, Personality, Experiment — for creating deeply contextual, high-quality prompts that produce exceptional content and research outputs. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Capacity/Role, Insight, Statement, Personality, Experiment Best for Content creation, research, communications Unique feature Experiment asks for multiple variations Compared to More comprehensive than RISEN for content **TL;DR:** CRISPE (Capacity/Role, Insight, Statement, Personality, Experiment) is a six-component framework that adds background insight and a defined personality to standard structured prompting, making outputs richer and more nuanced. Use it for complex content, research summaries, and communication tasks where context depth and tone precision both matter. Fill every component — skipping Personality or Insight produces generic results that miss CRISPE's main advantage. ## What is the CRISPE Framework? CRISPE is a structured prompt engineering framework developed to produce comprehensive, high-quality outputs for complex content, research, and communication tasks. It consists of six components that ensure the model has all necessary context before beginning its work: | Letter | Component | What it provides | | --- | --- | --- | | C | Capacity and Role | The model's expertise, identity, and capabilities for this task | | R | Role (in context) | Reinforces the specific perspective or authority the model brings | | I | Insight | Background context, data, or knowledge the model must consider | | S | Statement | The specific task, question, or request to be completed | | P | Personality | The tone, voice, style, or communication register to use | | E | Experiment | A request for multiple variations, approaches, or options | What distinguishes CRISPE from other frameworks is the explicit **Insight** component (background context that the model must internalize) and the **Personality** component (tone and voice specification). These two additions make CRISPE particularly powerful for content creation, communications, and research tasks where context and voice are as important as the instructions themselves. ## When to Use the CRISPE Framework 📰 ### Content Creation Blog posts, thought leadership articles, social media content — where voice, audience, and insight context all matter. 📢 ### Marketing Communications Email campaigns, product copy, press releases — where brand personality and market context must be encoded upfront. 🔬 ### Research Assistance Literature reviews, market research, expert analysis — where the model needs substantial background insight to reason correctly. 📝 ### Executive Communications Investor updates, board reports, executive summaries — high-stakes documents where tone, authority, and precision are non-negotiable. 🎨 ### Creative Projects Scripts, narratives, creative briefs — where multiple variations via the Experiment component help explore creative directions. 🎓 ### Educational Content Course materials, explainer content, learning resources — where the right pedagogical voice and learner context are essential. ## How to Build a CRISPE Prompt 1. 1 ### C/R — Define Capacity and Role Establish the model's expertise and authority for this task. Include specialization, experience level, and the type of organization they work for. This is the same as a detailed Role Prompt — it primes the model's knowledge and communication patterns. 2. 2 ### I — Provide Insight (background context) Share the context the model needs to do the task well: the company's positioning, the target audience's characteristics, key market facts, a style guide, or any data the model would otherwise have to guess. This is the component most commonly missing from naive prompts. 3. 3 ### S — State the task precisely Make the core request. Use a clear action verb and specify the deliverable, scope, and any structural requirements (sections, length, format). 4. 4 ### P — Define Personality Specify tone, voice, and style explicitly. "Authoritative but approachable", "conversational and energetic", "formal and precise" — these adjectives meaningfully shift the output. If you have a style guide, summarize its key principles here. 5. 5 ### E — Request Experiments (optional) Ask for 2–3 variations of the output if you want options to choose from. "Generate 3 versions: one that leads with data, one that leads with a story, and one that leads with a provocative question." The Experiment component turns one request into a creative exploration. ## Prompt Examples CRISPE — Thought Leadership Article ``` Capacity and Role: You are a veteran B2B SaaS marketing strategist with 15 years of experience building category-defining narratives for developer tools companies. You have written for publications like HackerNews, The Pragmatic Engineer, and company engineering blogs. Insight: Prompt Edit is a free macOS app that lets developers save and reuse AI prompt templates with dynamic variables. It stores everything locally (no cloud, no account). The target audience is developers and technical PMs who use AI daily and are frustrated by prompt repetition and lack of organization. The biggest competitor behavior is keeping prompts in Notion or text files. Statement: Write an 800-word thought leadership article titled "Why Your AI Prompts Deserve Version Control" for the Prompt Edit blog. Personality: Opinionated, direct, and slightly irreverent. Use developer analogies. Avoid marketing buzzwords and vague superlatives. Write as if explaining to a smart peer at a conference, not selling to them. Experiment: Provide 2 versions of the opening paragraph: Version A: starts with a frustrating scenario the reader recognizes Version B: starts with a bold, counterintuitive claim ``` CRISPE — Investor Update Email ``` Capacity and Role: You are an experienced startup communications advisor who has helped 50+ founders write investor updates. You write with clarity, honesty, and appropriate confidence. Insight: Monthly metrics: MRR $42k (+18% MoM), churn 3.2% (down from 4.1%), CAC $280, LTV $1,800. Hired 2 engineers. Launched enterprise tier. One major deal ($120k ARR) in final negotiation. Runway: 14 months. Statement: Write a monthly investor update email covering: metrics highlights, key wins, main challenge, and what help we need. Personality: Confident but not overconfident. Transparent about challenges without being alarming. Professional and concise. Founders who write updates their investors actually read. Experiment: Provide the challenge section in 2 tones: Version A: matter-of-fact and analytical Version B: solution-oriented and proactive ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Explicit Insight component prevents context-free outputs | 6 components make it heavier than simpler frameworks | | Personality component ensures brand voice consistency | Insight section can become very long for context-heavy tasks | | Experiment component generates creative options efficiently | Less suited for pure task execution vs RISEN's Step-focused approach | | Excellent for content, communications, and research tasks | Experiment adds token cost if you don't need multiple variations | ## CRISPE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Capacity and Role: Act as {{ role }}. Insight: {{ background_context }} Statement: {{ task }} Personality: Respond in the style of {{ voice_and_tone }}. Experiment: Give me {{ number_of_variations }} different versions so I can compare them. ``` ## Prompt Flows That Use CRISPE Framework CRISPE Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Test Strategist](https://www.promptedit.app/prompt-flow/the-test-strategist) · Software Development ## Frequently Asked Questions What is the CRISPE framework? CRISPE is a structured prompt engineering framework with six components: Capacity and Role (the model's expertise and identity), Insight (background context the model needs), Statement (the specific task or question), Personality (the tone, style, or voice to use), and Experiment (a request for multiple variations or options). Together, these produce highly contextual, well-calibrated outputs for complex content and research tasks. What does CRISPE stand for? CRISPE stands for: C — Capacity and Role, R — Role (combined with C), I — Insight, S — Statement, P — Personality, E — Experiment. The C and R are sometimes treated as one combined component defining who the model is and what expertise it has. How does CRISPE differ from RISEN? Both are structured prompt frameworks, but with different emphases. CRISPE includes Insight (background context) and Personality (tone/voice) as explicit first-class components, making it better suited for content creation, communications, and research assistance. RISEN is more task-execution-focused with Steps and explicit End Goal. CRISPE's Experiment component also uniquely asks for multiple output variations, which RISEN lacks. When should I use CRISPE? CRISPE excels for content creation (articles, social posts, scripts), research assistance, high-stakes communications (executive reports, investor updates), creative writing, and any task where tone, voice, and background context are as important as the task instructions. If you find yourself frequently specifying 'write this in the voice of...' or 'keep in mind that...', CRISPE formalizes those elements. What is the Experiment component in CRISPE? The Experiment component asks the model to produce multiple versions, variations, or approaches to your request. Instead of one answer, you get 2–3 options with different angles, tones, or approaches — allowing you to choose, combine, or use them as inspiration. This is particularly valuable for creative and communication tasks where there is no single 'correct' output. Can I skip some CRISPE components? Yes — treat CRISPE as a checklist of what to consider, not a rigid form to complete. For many tasks, Personality can be included in the Role definition, and Experiment is optional if you only need one version. The components that add the most value and are most often missing in naive prompts are Insight (context) and Personality (voice), so prioritize those. Is CRISPE good for technical prompts? CRISPE works for technical tasks but RISEN may be a better fit due to its explicit Steps component. For technical content creation (developer documentation, technical blog posts, architecture explainers), CRISPE's Personality and Insight components add meaningful value. For pure code generation or analysis tasks, RISEN or a simple Role + Instruction prompt is often more efficient. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/deep [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › D.E.E.P. Framework Iterative Prompting Intermediate # D.E.E.P. Framework Define the scope, Expand into a full draft, Evaluate critically, then Polish to publication quality — a four-stage iterative framework that takes AI-assisted content from raw idea to finished document. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Define-Expand-Evaluate-Polish Best for Research reports, white papers, long-form content Token cost Higher **TL;DR:** DEEP (Define, Expand, Evaluate, Polish) is an iterative four-stage framework that takes a content piece from scope definition through full drafting, critical self-evaluation, and final refinement. Use it for long-form documents — reports, articles, proposals — where a single-pass draft is rarely good enough. Run each stage as a separate prompt, feeding the previous output forward as input to the next stage. ## What is the D.E.E.P. Framework? The D.E.E.P. framework is an iterative prompt engineering approach that structures content creation as a four-stage progressive refinement process. Its stages — **Define**, **Expand**, **Evaluate**, and **Polish** — guide the AI through a disciplined workflow that mirrors how skilled human writers and researchers approach complex documents. - **D — Define:** Establish scope and success criteria - **E — Expand:** Generate the full draft - **E — Evaluate:** Critique against the definition - **P — Polish:** Refine to final form Unlike single-shot prompt frameworks that ask the AI to produce a finished output in one step, DEEP acknowledges that high-quality content cannot reliably emerge from a single generation event. By separating the work into four explicit stages with distinct objectives, DEEP dramatically increases the quality of final outputs — particularly for research synthesis, strategic analysis, white papers, and long-form editorial content. - **Define:** Establish the scope, angle, audience, and success criteria before any content is generated. - **Expand:** Generate the full draft based on the defined parameters. - **Evaluate:** Critically assess the draft against the definition criteria, identifying gaps, weaknesses, and improvement opportunities. - **Polish:** Refine the draft into final, publication-ready form based on the evaluation findings. ## When to Use the D.E.E.P. Framework 📋 ### Research Synthesis Define the research question and evaluation criteria, expand into a comprehensive synthesis, evaluate for gaps and bias, and polish into a structured research summary or literature review. 📑 ### White Papers and Reports Use DEEP's structured stages to take complex strategic or technical topics from rough outline through full draft to polished, publication-ready long-form documents. 📰 ### Long-Form Editorial Content Produce well-argued, nuanced articles and essays that go through genuine critical evaluation before reaching final form — not just a first-draft output with light editing. 📊 ### Strategic Memos and Analysis Define the decision criteria, expand the analysis, evaluate the rigor of each section, and polish into an executive document where every claim is supported and every recommendation is clearly argued. 🎓 ### Academic and Technical Writing Structure the Define stage around the research question and methodology, expand into a structured draft, evaluate against academic standards, and polish to meet journal or thesis requirements. 🤝 ### Collaborative Human-AI Editing Assign each DEEP stage to the appropriate collaborator — human strategist for Define, AI for Expand, human editor for Evaluate, AI for Polish — creating a structured editorial workflow. ## How to Use the D.E.E.P. Framework 1. 1 ### Define — Establish scope and success criteria Before any content is generated, define the document's purpose, audience, core argument or findings, and what a successful output looks like. This definition becomes the benchmark for the Evaluate stage. A weak Define stage produces a weak Evaluate stage because there are no agreed criteria to assess the draft against. Invest time here — it pays dividends in every subsequent stage. 2. 2 ### Expand — Generate the full draft Instruct the AI to produce a complete, unrestrained first draft based on the Define output. At this stage, completeness matters more than perfection. Ask the AI to cover all sub-topics, develop all arguments, and include all necessary sections — even if some will be revised or cut. A complete draft is easier to evaluate and refine than a cautious, abbreviated one. 3. 3 ### Evaluate — Critique against the definition Ask the AI (or perform yourself) a structured critique of the Expand draft using the Define criteria. Good evaluation prompts ask for: gaps in argumentation, unsupported claims, structural problems, unclear sections, tonal inconsistencies, and missed opportunities. The Evaluate stage should produce a prioritized list of specific improvements — not vague feedback like "make it more engaging". 4. 4 ### Polish — Refine to final form Apply the Evaluate findings to produce a final, publication-ready version. The Polish stage should address every specific improvement identified in Evaluate. Ask the AI to confirm which evaluation points have been addressed in the revision. A polished DEEP output should be indistinguishable from carefully edited professional writing. ## Prompt Examples White Paper Section on AI Adoption Barriers ``` Work through the following four stages to produce a white paper section on AI adoption barriers in mid-market financial services. DEFINE: Clarify the scope. The audience is CIOs and CTOs at banks and insurance companies with 200-2000 employees. The key argument is that the primary barrier is not technology cost but organizational readiness. Identify the three main sub-topics to cover. EXPAND: Write a full 500-word draft section covering all three sub-topics. Use a formal but accessible tone. Include at least two concrete examples or data references (you may use plausible illustrative statistics if real data is not available, clearly noted as illustrative). EVALUATE: Review the draft. Identify: (1) any logical gaps or unsupported claims, (2) any sections that repeat or could be tightened, (3) whether the organizational readiness argument is sufficiently foregrounded. POLISH: Rewrite the section incorporating all evaluation improvements. Ensure the final version flows as a single cohesive piece of professional writing with no evaluation notes visible. ``` Market Expansion Strategic Recommendation Memo ``` Use the DEEP framework to develop a strategic recommendation memo on whether our company should expand into the German market in the next 18 months. DEFINE: State the decision criteria that should drive the recommendation. Identify the 4 most important factors to analyze: market size potential, competitive landscape, regulatory complexity, and operational cost to enter. EXPAND: Write a structured memo draft with one section per factor. Each section should include an analysis statement, the key risk or opportunity, and a preliminary recommendation weight (high/medium/low consideration). EVALUATE: Assess the draft. Are the four factors treated with equal rigor? Is the overall recommendation clear, or does the memo avoid committing to a direction? What is the weakest section? POLISH: Revise into a final executive memo with a clear recommendation in the opening paragraph, supporting analysis in the body, and a single closing paragraph stating the proposed next step. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Produces significantly higher quality output than single-shot prompting | Higher token cost than single-shot frameworks | | Evaluation stage catches gaps before they reach the final draft | Requires more time and prompting effort per document | | Maps naturally to professional editorial and research workflows | Evaluate stage quality depends on the specificity of the Define stage | | Define stage prevents scope creep and misaligned output | Not well-suited for quick, short-form content requests | ## D.E.E.P. Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Define: {{ scope_and_success_criteria }} Expand: Write a full first draft of {{ deliverable }}. Evaluate: Critique the draft against the definition above and list every gap. Polish: Produce the final version that fixes those gaps. ``` ## Prompt Flows That Use D.E.E.P. Framework D.E.E.P. Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Brand Voice Engineer](https://www.promptedit.app/prompt-flow/the-brand-voice-engineer) · Brand Strategy ## Frequently Asked Questions What does D.E.E.P. stand for in prompt engineering? D.E.E.P. stands for Define, Expand, Evaluate, and Polish. It is an iterative prompt framework that guides both the AI and the human through a four-stage refinement process. Rather than expecting a perfect output from a single prompt, DEEP structures the work as progressive development: define the scope, expand into a full draft, evaluate critically, and polish to final quality. Why is DEEP called an iterative framework? DEEP is iterative because it treats content creation as a multi-stage process rather than a single generation event. Each stage — Define, Expand, Evaluate, Polish — produces output that feeds the next stage. You can run DEEP as a series of four separate prompts or as one structured mega-prompt that instructs the AI to work through all four stages in sequence. Either way, the process involves building on and critiquing previous output. When should I use the DEEP framework? Use DEEP for high-stakes content that requires genuine quality — research reports, white papers, long-form articles, strategic plans, and executive presentations. DEEP is the right choice when you cannot afford to publish a first draft, when the content will be reviewed by senior stakeholders, or when the topic is complex enough that a single prompt is unlikely to capture all the necessary nuance. Can I use DEEP in a single prompt or does it require multiple exchanges? DEEP can be used both ways. In a single-prompt approach, you instruct the AI to work through all four stages sequentially, explicitly completing each stage before proceeding to the next. In a multi-turn approach, you use a separate prompt for each stage, reviewing and adjusting between turns. The multi-turn approach gives you more control and the ability to redirect at each stage; the single-prompt approach is faster but less controllable. What happens in the Evaluate stage of DEEP? In the Evaluate stage, the AI — or you — critically reviews the expanded draft against the original definition criteria. A good Evaluate prompt asks the AI to identify gaps in the argument, check factual accuracy, assess whether the structure serves the reader, flag any sections that are unclear or unsupported, and suggest specific improvements before the Polish stage. Treating evaluation seriously is what separates DEEP outputs from ordinary first drafts. How is DEEP different from simply asking for a draft and then asking for revisions? Simple draft-then-revise prompting lacks the explicit Define and Evaluate stages that give DEEP its structure. In DEEP, Define ensures the scope, angle, and success criteria are agreed before any writing begins. Evaluate uses specific criteria — derived from the Define stage — to assess the draft. This prevents the common problem of revising in circles because the quality bar was never clearly established at the start. Is DEEP suitable for collaborative workflows with human editors? Yes, DEEP maps naturally onto collaborative workflows. The Define stage can be completed by a human strategist or editor. The Expand stage is delegated to the AI. The Evaluate stage is performed by a human editor using the AI's draft. The Polish stage can be a human-AI collaboration where the human provides specific revision notes and the AI implements them. DEEP's structure makes the human-AI collaboration explicit and organized. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/evoprompt [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › EvoPrompt Advanced Technique Advanced # EvoPrompt Apply evolutionary algorithms to prompt optimization — evolve a population of candidate prompts through fitness evaluation, crossover, and mutation to automatically discover high-performing prompts. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Guo et al., 2023 (arXiv:2309.08532) Key innovation Black-box evolutionary search over prompt space Algorithms Genetic Algorithm (GA), Differential Evolution (DE) Best for Automated optimization, research, production scale **TL;DR:** EvoPrompt applies evolutionary algorithms — selection, crossover, mutation — to automatically optimize prompts across generations, using an LLM to perform the crossover and mutation steps intelligently. Use it when you have a measurable task metric and need to find a high-performing prompt without manual trial-and-error. Start with a diverse initial population of 5-10 candidate prompts and let the algorithm evolve them over 10-20 generations guided by objective performance scores. ## What is EvoPrompt? EvoPrompt is a research technique introduced by Guo et al. in 2023 that applies evolutionary computation to the problem of automatic prompt optimization. Instead of having a human iterate on prompt variants manually, EvoPrompt maintains a population of candidate prompts and uses evolutionary operators — selection, crossover, and mutation — to search the space of possible prompts guided by objective task performance. The critical insight is that the crossover and mutation operations are themselves performed by an LLM: rather than splicing text randomly (as in traditional genetic algorithms), EvoPrompt prompts an LLM to intelligently combine or modify candidate prompts in semantically meaningful ways. This makes the search space exploration far more efficient than random perturbation. The algorithm operates in three phases, repeated for N generations: - **Evaluation (Fitness):** Each candidate prompt in the population is run against a labeled development set. Task performance (accuracy, F1, etc.) becomes the fitness score. - **Selection:** The highest-fitness candidates are selected as parents for the next generation. Poor performers are discarded. - **Variation (Crossover + Mutation):** An LLM combines pairs of selected prompts (crossover) and makes targeted modifications to individual prompts (mutation) to produce new offspring candidates for the next generation. ## When to Use EvoPrompt 🔬 ### Research & Benchmarking EvoPrompt was designed for research settings where you need to find near-optimal prompts for a task to establish a performance baseline — without the bias of human prompt intuition. 🏭 ### High-Volume Production Tasks When a prompt runs millions of times, even a 3–5% accuracy improvement from evolved prompts translates to enormous value. The upfront optimization cost pays off quickly at scale. 🏷️ ### Classification & Labeling Pipelines Tasks with clear objective metrics (accuracy, F1) are ideal for EvoPrompt — the fitness function is unambiguous and the evolutionary search has a clear signal to follow. 🌐 ### Cross-Lingual Task Optimization EvoPrompt can simultaneously optimize prompt phrasing for multiple languages, discovering language-specific phrasings that outperform direct translations of English prompts. 📊 ### Novel Task Domains For tasks where human prompt engineering intuition is weak — specialized scientific domains, rare language phenomena, unusual output formats — evolutionary search can discover effective approaches humans wouldn't think to try. ⚙️ ### Model-Agnostic Prompt Transfer EvoPrompt can re-optimize a prompt that works well on one model for a different model, finding the optimal phrasing for the target model's specific sensitivities. ## How to Implement EvoPrompt 1. 1 ### Define the task and fitness function Specify what task you are optimizing for and how to measure prompt performance. For classification, this is typically accuracy on a held-out development set of 50–200 labeled examples. The fitness function must be computable automatically — EvoPrompt runs dozens of evaluation cycles and cannot rely on human judgment for each. 2. 2 ### Initialize a diverse population of seed prompts Create 10–20 candidate prompts with significant structural and stylistic diversity. Include: direct task instructions, persona-based prompts, few-shot example prompts, chain-of-thought prompts, and format-specific prompts. Diversity in the initial population is essential — EvoPrompt cannot explore what is not represented in the gene pool. 3. 3 ### Evaluate fitness for each candidate Run every candidate prompt on all development set examples and compute the fitness score. This is computationally expensive — with 20 candidates and 100 dev examples, that is 2,000 LLM calls per generation. Batch API calls and use smaller, faster models for evaluation if budget is constrained. 4. 4 ### Select parents and apply crossover Select the top N candidates by fitness. For each new offspring, sample two parents and prompt an LLM to produce a crossover: "Combine the strongest elements of these two prompts to create a new variant that may perform better on [task]." The LLM performs semantically intelligent recombination rather than random text splicing. 5. 5 ### Apply mutation to selected candidates Apply mutation to a subset of the population by prompting an LLM to make 1–3 targeted modifications to a candidate prompt (rephrase an instruction, add a constraint, change the output format, etc.). Mutation maintains diversity and prevents premature convergence to a local optimum. 6. 6 ### Repeat for N generations and select the best Run 5–20 generations depending on budget. Track the best fitness score achieved across all generations. When improvement plateaus for 3+ generations, stop. Validate the best evolved prompt on a held-out test set (not the dev set used for fitness evaluation) to confirm the improvement generalizes. ## Prompt Examples EvoPrompt Generation Cycle — Crossover Example ``` ## EvoPrompt — Manual Simulation (Generation 1) ### Task Sentiment classification: classify customer reviews as Positive, Negative, or Neutral. ### Initial Population (3 candidates for illustration) **Candidate A:** Classify the sentiment of the following review as Positive, Negative, or Neutral. Review: {review} Sentiment: **Candidate B:** You are a sentiment analysis expert. Read the review carefully and determine whether the overall customer sentiment is Positive, Negative, or Neutral. Consider the overall impression, not just individual words. Review: {review} Answer with one word: **Candidate C:** Task: Sentiment Analysis Instructions: Analyze the emotional tone of the customer review below. Output exactly one of: Positive | Negative | Neutral Do not explain your answer. Review: {review} Sentiment: ### Fitness Evaluation (accuracy on 50-example dev set) - Candidate A: 74% - Candidate B: 81% - Candidate C: 78% ### Crossover Operation (LLM combines B and C) Prompt the LLM: "Below are two high-performing prompt variants for a sentiment classification task. Create a new improved variant that combines the strongest elements of both. Keep the best structural and instructional elements from each." [Insert Candidate B and Candidate C] ### Mutated Offspring (Generation 2 candidate) You are a sentiment analysis expert. Analyze the emotional tone of the customer review below and determine whether the overall impression is Positive, Negative, or Neutral. Consider the full context, not isolated words. Output exactly one of: Positive | Negative | Neutral Review: {review} Sentiment: ``` Mutation Operator Template ``` ## EvoPrompt — Mutation Prompt Template You are a prompt optimization assistant implementing the mutation operation of an evolutionary algorithm for prompt optimization. Below is a prompt that achieves {fitness_score}% accuracy on a {task_name} task. Your goal is to produce a MUTATED variant that may perform better. Mutation guidelines: - Make 1–3 targeted changes, not a complete rewrite - Valid mutations include: rephrasing an instruction, adding a constraint, changing the output format specification, adding or removing an example, adjusting the persona or tone, reordering sections - Preserve elements that are likely contributing to performance - Introduce one novel element not present in the original Original prompt: --- {original_prompt} --- Output ONLY the mutated prompt text. No explanation, no commentary. The mutated prompt should be immediately usable as a replacement. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Discovers prompt phrasings that human engineers would not think to try | Very high computational cost — thousands of LLM calls per optimization run | | Fully black-box — works with any LLM API without gradient access | Requires a labeled development set with a clear, automatable fitness function | | Objective optimization guided by task performance, not human preference | Evolved prompts may overfit to the dev set and not generalize | | Scales to large evaluation sets for statistically robust optimization | Impractical for low-volume tasks where the optimization cost exceeds the benefit | ## EvoPrompt Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` You are optimising a prompt for this task: {{ task_description }} Parent prompt A (score {{ score_a }}): """ {{ prompt_a }} """ Parent prompt B (score {{ score_b }}): """ {{ prompt_b }} """ 1. Crossover: combine the strongest elements of A and B into one prompt. 2. Mutation: make one deliberate change (wording, structure or constraint) to the result. 3. Return only the new child prompt. ``` ## Frequently Asked Questions What is EvoPrompt? EvoPrompt (Evolutionary Prompt Optimization) is a research technique (Guo et al., 2023) that applies evolutionary algorithms to automatically optimize LLM prompts. It maintains a population of candidate prompts, evaluates each on a task performance metric, then applies genetic operations — crossover (combining elements of high-performing prompts) and mutation (making small variations) — to produce improved candidate prompts. The cycle repeats for multiple generations without any gradient computation. How does EvoPrompt differ from manual prompt engineering? Manual prompt engineering relies on a human iterating on one or a few prompt variants based on qualitative intuition. EvoPrompt systematically explores a much larger space of prompt variants in parallel, guided by objective task performance rather than human judgment. It can discover prompt phrasings and structures that humans would be unlikely to try, and it scales to large test sets where manual evaluation would be infeasible. What evolutionary algorithms does EvoPrompt use? The original paper (Guo et al., 2023) implements two evolutionary strategies: Genetic Algorithm (GA)-based EvoPrompt and Differential Evolution (DE)-based EvoPrompt. GA-EvoPrompt uses selection, crossover between pairs of parents, and random mutation. DE-EvoPrompt uses a differential perturbation strategy where a new candidate is generated by adding the difference between two population members to a base prompt. Both are guided by an LLM that performs the actual crossover and mutation operations using natural language. Does EvoPrompt require access to model gradients or weights? No. This is EvoPrompt's key advantage for practical use: it operates entirely as a black-box optimizer. It only requires the ability to run a prompt against a task and measure performance. This means it works with closed-source model APIs (GPT, Claude, Gemini) where gradient information is unavailable, unlike gradient-based prompt optimization methods like AutoPrompt or Prefix Tuning. What is the fitness function in EvoPrompt? The fitness function is the performance metric that evaluates how well each candidate prompt performs on the target task. For classification tasks, this is typically accuracy on a development set. For generation tasks, it might be ROUGE score, BERTScore, or a human preference score. The fitness function is task-specific and must be defined before running EvoPrompt — it is the objective that the evolutionary search is optimizing toward. How large should the initial prompt population be? The original paper uses populations of 10–20 candidate prompts. Too small a population risks premature convergence to a local optimum. Too large increases evaluation cost (each prompt must be run on the dev set for scoring). A practical starting point is 10 candidates for constrained budgets and 20–30 for research or high-stakes production optimization. Initial diversity is important — seed the population with prompts that vary in structure, length, and phrasing strategy. When is EvoPrompt worth the computational cost? EvoPrompt is worth the cost when: (1) you have a high-volume production task where even a 5% accuracy improvement has significant business value, (2) you have a well-defined evaluation metric and a labeled development set, (3) the task is novel enough that human prompt intuition is unreliable, or (4) you are conducting research on prompt optimization. For low-volume or exploratory tasks, manual refinement with a Prompt Refinement Loop is more cost-effective. ## Origin & Sources 1. Guo et al. (2023). [Connecting Large Language Models with Evolutionary Algorithms Yields Powerful Prompt Optimizers](https://arxiv.org/abs/2309.08532) . arXiv / ICLR 2024. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/few-shot [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Few-Shot Prompting In-Context Learning Beginner # Few-Shot Prompting Provide a handful of input-output examples directly in your prompt so the AI learns the exact pattern, format, or style you need — no model fine-tuning required. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as In-Context Learning, k-Shot Learning Origin Brown et al., OpenAI GPT-3 (2020) Typical examples 2–8 input-output pairs Token cost Moderate (examples consume tokens) **TL;DR:** Few-Shot Prompting includes 2-8 worked input-output examples in your prompt so the model can infer the exact pattern you need without any fine-tuning. Use it when zero-shot instructions produce inconsistent formats, wrong styles, or niche outputs the model doesn't handle reliably by default. Place examples before your actual request and ensure they cover the range of variation the model will encounter. ## What is Few-Shot Prompting? Few-Shot Prompting (also called in-context learning) is the practice of including 2 to 8 worked examples directly in your prompt context before making your actual request. These examples show the model exactly what a correct input-output pair looks like, enabling it to infer the task pattern without any model fine-tuning. The concept was popularized in the GPT-3 paper by Brown et al. (2020), which demonstrated that large language models could learn from examples purely in context — a form of "learning without learning", since no weights are updated. There are three variants by example count: - **One-Shot:** Exactly one example. Useful when a single demonstration is enough to establish the format. - **Few-Shot:** 2–8 examples. The most practical range for most tasks. - **Many-Shot:** 10+ examples. Helpful for complex tasks where diversity of coverage matters more than token efficiency. ## When to Use Few-Shot Prompting 🏷️ ### Classification & Labeling Sentiment analysis, intent detection, topic categorization — showing example label assignments dramatically improves consistency. 📝 ### Structured Output JSON, XML, CSV, or any specific format where the model must produce machine-parseable output without deviation. 🌍 ### Translation & Localization Show the model examples in your target dialect, formality level, or specialized domain vocabulary before the actual translation request. ✍️ ### Style & Tone Matching Provide 3–5 examples of your brand voice, writing style, or persona before asking for new content in the same style. 🔄 ### Data Transformation Reformatting, extraction, normalization — show the model the input format and desired output format with 2–3 pairs. 🧩 ### Specialized Domain Tasks Legal clause extraction, medical coding, financial analysis — use examples to teach domain-specific conventions the model may not know reliably. ## How to Use Few-Shot Prompting 1. 1 ### Define the task format precisely Decide on the exact input format and desired output format. Consistency between examples is critical — any variation confuses the model about which variation to follow. 2. 2 ### Select 2–5 diverse, representative examples Choose examples that cover different sub-cases (e.g., all sentiment classes if doing sentiment analysis). Avoid examples that are too similar or that represent only the easy cases. 3. 3 ### Format examples consistently Use a clear, consistent delimiter between input and output. Common formats: "Input: ... Output: ...", "Q: ... A: ...", or a labeled JSON pair. The model will follow whichever pattern you establish. 4. 4 ### End with the actual request (incomplete) After your examples, provide the actual input and the beginning of the output label, leaving the answer for the model to complete. For example: "Input: [new input]\nOutput:" — this signals the model to continue the pattern. ## Prompt Examples Sentiment Classification ``` Classify the sentiment of each review as Positive, Negative, or Neutral. Review: "The delivery was fast and the packaging was perfect." Sentiment: Positive Review: "Product broke after one week. Very disappointed." Sentiment: Negative Review: "It's okay, does what it says on the box." Sentiment: Neutral Review: "Absolutely love this — best purchase I've made this year!" Sentiment: ``` Structured JSON Extraction ``` Extract the key details from each job posting as JSON. Posting: "Senior React Developer at Acme Corp. Remote. $120k–$150k/year." Output: {"title": "Senior React Developer", "company": "Acme Corp", "location": "Remote", "salary": "$120k–$150k"} Posting: "Marketing Manager – London office. Full-time. Competitive salary." Output: {"title": "Marketing Manager", "company": null, "location": "London", "salary": "Competitive"} Posting: "Data Analyst at FinTech Startup Berlin. Hybrid. €65k–€80k." Output: ``` Brand Voice Style Matching ``` Rewrite each sentence in our brand voice: friendly, concise, and action-oriented. Original: "Our platform enables users to perform data synchronization operations." Rewritten: "Sync your data in seconds — effortlessly." Original: "The application facilitates the management of team-based tasks." Rewritten: "Keep your team's work organized and moving forward." Original: "Users are able to configure notification preferences within the settings menu." Rewritten: ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dramatically improves format and style consistency | Examples consume tokens, increasing cost and latency | | No model fine-tuning required — works at inference time | Poorly chosen examples can introduce bias | | Works with any large language model | Context window limits cap how many examples you can include | | Examples can be swapped easily without retraining | For very large example sets, fine-tuning is more efficient | ## Few-Shot Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` {{ task_instruction }} Example 1 Input: {{ example_input_1 }} Output: {{ example_output_1 }} Example 2 Input: {{ example_input_2 }} Output: {{ example_output_2 }} Example 3 Input: {{ example_input_3 }} Output: {{ example_output_3 }} Now do the same for: Input: {{ new_input }} Output: ``` ## Prompt Flows That Use Few-Shot Prompting Few-Shot Prompting is one step in these 3 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Brand Voice Engineer](https://www.promptedit.app/prompt-flow/the-brand-voice-engineer) · Brand Strategy - [The Onboarding Writer](https://www.promptedit.app/prompt-flow/the-onboarding-writer) · Documentation - [The Test Strategist](https://www.promptedit.app/prompt-flow/the-test-strategist) · Software Development ## Frequently Asked Questions What is Few-Shot Prompting? Few-Shot Prompting is a technique where you include 2 to 8 worked input-output examples directly in the prompt before your actual request. These examples act as demonstrations that teach the model the exact pattern, format, tone, or reasoning style you expect — without updating the model's weights. It is a form of in-context learning. How many examples should I include in a Few-Shot prompt? Typically 2–5 examples strike the best balance. More examples improve accuracy but consume more tokens and can push earlier context out of the model's attention window. Research shows diminishing returns beyond 8–10 examples for most tasks. Start with 3 examples and increase if results are inconsistent. What is the difference between One-Shot, Few-Shot, and Zero-Shot prompting? Zero-Shot provides no examples and relies entirely on instructions. One-Shot provides exactly one example. Few-Shot provides 2 or more examples. One-Shot is a special case of Few-Shot. As you add more examples, the model better understands the desired output pattern, at the cost of more tokens. When does Few-Shot Prompting outperform Zero-Shot? Few-Shot outperforms Zero-Shot when you need a very specific output format, for classification or labeling tasks with a fixed set of labels, when the task is niche or unusual (outside common training distributions), and when tone, register, or style consistency is critical. How should I select good examples for Few-Shot prompting? Choose examples that are diverse (cover different sub-cases), representative (reflect real inputs the model will encounter), unambiguous (clear correct answer), and roughly the same length as the target input. Avoid examples that are too similar to each other or that introduce edge-case reasoning the model might generalize incorrectly. Can Few-Shot Prompting introduce bias? Yes. If your examples skew toward a particular label, pattern, or style, the model will bias its outputs accordingly. For example, if all your classification examples return 'positive', the model may over-predict 'positive'. Always balance examples across all expected output classes. Does Few-Shot Prompting work with any AI model? Few-Shot Prompting works with virtually all modern language models that support long context inputs. Larger, more capable models extract more signal from examples. Smaller models may struggle to generalize from examples if the task is complex. Always test with your target model. ## Origin & Sources 1. Brown et al. (2020). [Language Models are Few-Shot Learners](https://arxiv.org/abs/2005.14165) . arXiv / NeurIPS 2020. 2. Anthropic (2026). [Prompt engineering overview](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview) . Claude Docs. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/flow [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › FLOW Framework Structured Framework Intermediate # FLOW Framework Define not just what you want but what great looks like — the FLOW framework combines Function, Level, Output, and a Win Metric to anchor every AI task in an explicit quality standard. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Function · Level · Output · Win Metric Signature feature Win Metric defines the quality bar Best for Business analysis, optimization, technical work Complexity Medium — 4 focused components **TL;DR:** FLOW (Function, Level, Output, Win Metric) is a four-component framework that anchors every prompt in a measurable success criterion — the Win Metric — so the AI knows exactly what "done well" means before it starts. Use it for business deliverables where quality is judged against an objective standard: proposals, analyses, and content with specific performance goals. Define the Win Metric as a concrete, verifiable outcome rather than a vague quality description. ## What is the FLOW Framework? FLOW is a prompt engineering framework built around four components: **Function**, **Level**, **Output**, and **Win Metric**. Its central innovation is the Win Metric — an explicit definition of what a successful output looks like, stated before the AI begins work. - **F — Function:** What Should the AI Do? - **L — Level:** At What Expertise and Complexity? - **O — Output:** Define the Exact Deliverable Format - **W — Win Metric:** Define What Success Looks Like Most prompts tell the AI what to do but leave quality judgment entirely to the model. FLOW closes that gap by asking you to articulate the real-world test your output must pass: Can a CFO present it in a board meeting? Does a junior developer implement it without asking follow-up questions? Does the query run in under 30 seconds? When you define the win condition upfront, the AI can calibrate toward it. The Level component adds a second dimension of control: you specify not just the topic but the expertise level and complexity tier of the response, ensuring the output is right for your actual audience rather than a generic reader. ## When to Use the FLOW Framework 📈 ### Business Analysis Define the function (analyse unit economics), the audience level (senior management), and the win condition (presentable to a board) to get analysis that actually drives decisions. ⚡ ### Performance Optimisation Set a measurable win metric for technical optimisation tasks — query execution time, load speed, memory usage — so the AI targets a real benchmark, not a vague improvement. 🔍 ### Code Review Specify the function (review this PR), the level (senior engineer standard), and a win metric (a junior developer can action every comment without clarification). 📄 ### Executive Deliverables Produce strategy memos, board decks, and executive summaries calibrated to the specific expertise level and time constraints of senior leadership. 🏗️ ### Technical Documentation Specify the audience level and define the win condition (a new hire can onboard using this doc alone) to ensure documentation is complete and appropriately detailed. 🎯 ### Conversion & Copy Optimisation Set copy function and audience level, then define the win metric as a specific conversion outcome or user action — keeping creative work grounded in business objectives. ## How to Use the FLOW Framework 1. F ### Function — What Should the AI Do? State the primary function clearly and specifically. "Analyse", "optimise", "write", "review", "summarize" — be precise about the action. Include any key inputs the AI needs to perform the function. This is your instruction, not your goal. 2. L ### Level — At What Expertise and Complexity? Specify the expertise tier of the response and the assumed knowledge level of the reader. This could be a role ("senior management consultant"), an education level ("assumes MBA-level financial literacy"), or a complexity instruction ("explain without jargon, suitable for a non-technical executive"). 3. O ### Output — Define the Exact Deliverable Format Describe the structure, format, and sections of the output. Use numbered lists, specify tables, name sections, and set length expectations. A precise Output spec prevents the AI from choosing an arbitrary format that doesn't fit your workflow. 4. W ### Win Metric — Define What Success Looks Like State the real-world test your output must pass. Frame it in terms of who will use the output and what they need to do with it. The Win Metric is the quality bar — concrete, verifiable, and audience-specific. It is the most important component for high-stakes deliverables. ## Prompt Examples Strategy Memo — Unit Economics Analysis ``` Function: Analyse the unit economics of a direct-to-consumer subscription box business and identify the three biggest levers for improving profitability. Level: Senior management consultant at a top-tier strategy firm. Assume the reader has an MBA and is fluent in financial modelling, LTV/CAC ratios, contribution margins, and cohort analysis. Output: A structured two-page memo with: 1. Executive summary (3–4 sentences) 2. Current unit economics snapshot (table format) 3. Three profitability levers, each with: diagnosis, root cause, and recommended action 4. Prioritisation matrix (impact vs. effort, 2x2) Win Metric: The memo is successful if a CFO could read it in under 5 minutes, immediately understand the top priority, and walk into a board meeting ready to present the recommendation with confidence. ``` SQL Query Optimisation ``` Function: Optimise the following SQL query for a PostgreSQL 15 database. The query runs a daily report on a 50-million-row orders table and currently takes 4 minutes 20 seconds to execute. [Paste query here] Level: Database performance engineer with deep expertise in PostgreSQL query planning, indexing strategies, partition pruning, and EXPLAIN ANALYSE output interpretation. Write explanations at a level appropriate for a senior backend engineer who understands SQL but is not a DBA specialist. Output: 1. Annotated EXPLAIN ANALYSE breakdown of the current bottlenecks 2. Optimised version of the query with inline comments explaining each change 3. Index recommendations (DDL statements ready to run) 4. Expected performance improvement and reasoning Win Metric: The optimised query should run in under 30 seconds on equivalent hardware. Each recommendation must include the specific bottleneck it addresses so the engineer can validate the fix independently. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Win Metric anchors the AI to a concrete, verifiable quality standard | Requires upfront thinking to articulate a meaningful Win Metric | | Level component ensures output is calibrated for the actual audience | Less suited for exploratory or open-ended creative tasks | | Output spec prevents arbitrary formatting choices | Four components can be over-engineering for simple, low-stakes requests | | Excellent for business-critical and technically complex deliverables | | ## FLOW Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Function: {{ what_the_ai_should_do }} Level: {{ expertise_and_complexity_level }} Output: {{ exact_deliverable_format }} Win metric: {{ what_success_looks_like }} ``` ## Frequently Asked Questions What does FLOW stand for? FLOW stands for Function, Level, Output, and Win Metric. Function defines what the AI should do, Level sets the expertise and complexity tier of the response, Output specifies the exact deliverable format, and Win Metric defines how success will be judged — making the quality bar explicit before the AI begins. What makes FLOW different from other frameworks? FLOW's defining feature is the Win Metric — a concrete definition of what a successful output looks like. Instead of leaving quality judgment to the AI, you explicitly state the standard the output must meet. This is especially valuable for performance-critical, business, or technical tasks where vague outputs are costly. When should I use the FLOW framework? Use FLOW for performance-oriented, business-critical, or technically complex tasks where quality matters as much as completion: strategic memos, database optimisation, code reviews, business analysis, and any deliverable where you can articulate a clear success standard upfront. How do I write a good Win Metric? A good Win Metric describes the real-world test your output must pass. Think about who will use the output and what they need to do with it: 'A CFO could present this in 5 minutes', 'A junior developer can implement this without asking questions', 'The query runs in under 30 seconds'. Concrete, role-based tests work best. What should the Level component specify? Level should specify both the expertise level of the response (beginner, intermediate, expert, executive) and any complexity constraints (technical depth, vocabulary assumed, length expectations). It ensures the output is calibrated for your actual audience rather than defaulting to generic explanations. Can FLOW be used for creative tasks? Yes, with adaptation. For creative work, the Win Metric becomes qualitative: 'The copy should feel like Patagonia's brand voice', 'The story should make the reader feel urgency without panic'. As long as you can describe what success feels like, FLOW's structure keeps creative briefs anchored in clear expectations. How does FLOW compare to the SMART framework? Both FLOW and SMART emphasise measurable outcomes. SMART comes from project management goal-setting and works well for broad task scoping. FLOW is more granular — it focuses on the deliverable format and quality bar rather than the overall goal. They can be used together: SMART to define the project goal, FLOW to specify each AI-generated deliverable within it. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/focus [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › FOCUS Framework Structured Framework Intermediate # FOCUS Framework The only structured framework with an Underlying Assumptions component — FOCUS surfaces hidden context that would otherwise silently invalidate strategic and analytical outputs. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Function · Outcome · Criteria · Underlying Assumptions · Strategy Best for Strategic analysis, decision-making, consulting Unique feature Underlying Assumptions makes hidden context explicit **TL;DR:** FOCUS (Function, Outcome, Criteria, Underlying Assumptions, Strategy) is a strategic analysis framework that surfaces hidden assumptions before analysis begins, preventing advice built on unstated premises. Use it for consulting deliverables, strategic decisions, and business recommendations where flawed assumptions are the most common failure mode. Explicitly list 2-3 Underlying Assumptions and invite the model to challenge them if they appear incorrect. ## What is the FOCUS Framework? FOCUS is a structured prompt engineering framework built for strategic analysis, complex decision-making, and consulting-style outputs. The acronym stands for **Function, Outcome, Criteria, Underlying Assumptions, and Strategy** — five components designed to make every relevant constraint, expectation, and hidden belief explicit before the model begins its analysis. - **F — Function:** Define the role and task type - **O — Outcome:** State the desired end result - **C — Criteria:** Set measurable success standards - **U — Underlying Assumptions:** Surface hidden context - **S — Strategy:** Specify the analytical approach The framework's most distinctive feature is the **Underlying Assumptions** component. In most analytical prompts, critical context — budget constraints, team capabilities, market conditions, technical limitations — goes unstated because the prompter assumes the model will infer it correctly. FOCUS treats these assumptions as a first-class element of the prompt, forcing you to surface them deliberately. This matters because a strategically incorrect answer — one that is well-reasoned but based on wrong assumptions — is often worse than no answer at all. FOCUS is designed for situations where the quality of reasoning matters as much as the correctness of the conclusion. ## When to Use the FOCUS Framework 📈 ### Strategic Analysis Market entry, competitive positioning, growth strategy — any analysis where the conclusion depends heavily on assumptions about the competitive landscape or internal capabilities. ⚖️ ### Decision-Making Build/buy/partner decisions, technology choices, organizational structure changes — where trade-offs are complex and wrong assumptions lead to wrong choices. 🏛️ ### Consulting Deliverables Feasibility assessments, risk analyses, and strategic recommendations where the output must be grounded in the client's actual constraints, not theoretical best practices. 🏗️ ### Architecture Reviews Technical architecture critique and recommendations where team maturity, operational constraints, and business requirements shape what is actually viable. 💼 ### Business Case Development Investment proposals and business cases where financial assumptions, market assumptions, and execution assumptions must be made transparent to be credible. 🔬 ### Policy & Regulatory Analysis Compliance assessments and policy impact analyses where jurisdictional assumptions, timeline assumptions, and scope assumptions determine the entire framing. ## How to Use the FOCUS Framework 1. 1 ### Function — Define the role and task type Specify what the model is being asked to do and in what capacity. "Act as a strategic analyst performing a market entry assessment" is far more directive than "analyze this market." Function sets the operating mode and expertise level for the entire response. 2. 2 ### Outcome — State the desired end result Define the concrete deliverable you want — a recommendation, a risk register, a comparison table, a decision framework. Be specific about the form the output should take, not just the topic it should cover. 3. 3 ### Criteria — Set measurable success standards Define what a successful response looks like in concrete terms: how many risks to identify, what format for recommendations, which dimensions to cover. Criteria transform subjective quality into objective completeness. 4. 4 ### Underlying Assumptions — Surface hidden context List every assumption your analysis rests on: financial constraints, team capabilities, timeline, market conditions, technical limitations, regulatory status. If the model gets these wrong, every conclusion that follows is built on a false foundation. Write them explicitly — even the ones that seem obvious to you. 5. 5 ### Strategy — Specify the analytical approach Tell the model what kind of thinking to apply: first-principles reasoning, benchmarking, risk-weighted analysis, scenario planning. Strategy prevents the model from defaulting to a generic framework when a specific analytical lens is more appropriate for your situation. ## Prompt Examples Business Strategy — European Market Entry Assessment ``` Function: Perform a strategic market entry feasibility assessment. Outcome: A structured assessment of whether our SaaS project management tool should expand into the European SMB market in the next 12 months, with a clear recommendation. Criteria: The assessment must cover market size, competitive landscape, regulatory considerations (GDPR), and go-to-market channel options. It should conclude with a binary recommendation (enter now / defer) and three supporting rationale points for whichever position is taken. Underlying Assumptions: We assume our current ARR is $2.4M, growing at 40% YoY. We assume our product is English-only with no localization budget in the next 6 months. We assume GDPR compliance is not yet certified but is achievable within 4 months. We assume we have one dedicated sales hire available for European outreach. Strategy: Apply a risk-weighted feasibility lens rather than an opportunity-maximization lens — we need to know what could go wrong and how likely those risks are, more than we need to know the theoretical upside. ``` Engineering — Microservices Migration Architecture Review ``` Function: Act as a senior engineering architect reviewing a proposed microservices migration plan. Outcome: An architectural critique of the proposed migration from a monolith to microservices, identifying critical risks and recommending a phased approach if warranted. Criteria: Identify at least 4 architectural risks. For each risk, provide a severity rating (High / Medium / Low) and a specific mitigation strategy. If a phased approach is recommended, define the phases and their sequencing rationale. Underlying Assumptions: We assume the current monolith is a 6-year-old Rails application with no automated test coverage above 30%. We assume the team is 8 engineers, none of whom have built microservices before. We assume the business cannot tolerate more than 4 hours of planned downtime per quarter. We assume Kubernetes is the target deployment platform but the team has no production Kubernetes experience. Strategy: Prioritize operational and team-capability risks over theoretical architectural purity. The goal is a realistic plan this team can execute, not the ideal architecture in the abstract. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Underlying Assumptions component prevents conclusions built on wrong foundations | Requires significant upfront thinking to enumerate assumptions correctly | | Criteria component makes quality measurable rather than subjective | Overkill for simple questions where assumptions are not a meaningful risk | | Strategy component directs the model's analytical approach, not just its output | The Underlying Assumptions section can become unwieldy for highly complex scenarios | | Produces outputs that are grounded in real-world constraints rather than theoretical ideals | | ## FOCUS Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Function: {{ role_and_task_type }} Outcome: {{ desired_end_result }} Criteria: {{ success_standards }} Underlying assumptions: {{ assumptions }} Strategy: {{ analytical_approach }} ``` ## Prompt Flows That Use FOCUS Framework FOCUS Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Debate Judge](https://www.promptedit.app/prompt-flow/the-debate-judge) · Decision-Making ## Frequently Asked Questions What does FOCUS stand for in prompt engineering? FOCUS stands for Function, Outcome, Criteria, Underlying Assumptions, and Strategy. It is a five-component framework built for strategic and analytical prompts where making hidden context explicit is as important as the question itself. What makes the 'Underlying Assumptions' component unique? Underlying Assumptions is the component that distinguishes FOCUS from other frameworks. It forces you to surface the hidden beliefs, constraints, or conditions your analysis rests on — things like 'we assume the market is growing at 8% per year' or 'we assume the team has no machine learning expertise'. When these go unstated, the model may produce a technically correct answer that is practically useless. Making assumptions explicit grounds the output in reality. When should I use FOCUS over GUIDE or QUEST? Use FOCUS when the quality of the answer depends heavily on unstated context — strategic decisions, consulting analyses, and complex trade-off evaluations. GUIDE is better when you need structured expert advice with an evaluation loop. QUEST is better when you need scoped research. FOCUS is the strongest choice when you are worried that wrong assumptions could invalidate the entire answer. What goes in the 'Function' component? Function describes the task or role the model is performing — what it is being asked to do. Think of it as the job title for this specific prompt: 'Act as a strategic analyst', 'Function as a code reviewer', or 'Perform a market feasibility assessment'. It sets the operating mode for everything that follows. How do I write effective success Criteria? Criteria should be as concrete and measurable as possible. Instead of 'give me a good analysis', write 'The analysis should identify at least 3 strategic risks, rank them by severity, and provide one mitigation strategy per risk.' Vague criteria produce vague responses. The more specific your criteria, the more consistently useful the output will be. Can FOCUS be used for creative tasks? FOCUS can be adapted for creative work, but it is most powerful in analytical and strategic contexts. For purely creative tasks — copywriting, fiction, design briefs — the IDEA framework's iterative structure is a better fit. For creative-strategic hybrids like brand positioning or messaging strategy, FOCUS excels because assumptions about the audience and market heavily influence what 'creative' means in that context. How does Strategy differ from Direction in the GUIDE framework? Strategy in FOCUS defines the high-level approach or methodology for solving the problem — the 'what kind of thinking' rather than the step-by-step sequence. Direction in GUIDE is more procedural, specifying the order of steps. FOCUS Strategy might say 'Apply a first-principles analysis rather than a benchmarking approach', whereas GUIDE Direction might say 'First identify the root cause, then propose three ranked solutions.' ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/grow [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › GROW Framework Coaching Framework Beginner # GROW Framework The world's leading coaching model — Goal, Reality, Options, Will — adapted for AI prompts that guide clear thinking from your honest current situation to a committed, specific action plan. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Goal · Reality · Options · Will (Way Forward) Best for Coaching conversations, goal setting, problem solving, career planning Complexity Medium — requires honest self-assessment for best results Token cost Medium to High **TL;DR:** GROW (Goal, Reality, Options, Will) is a coaching model that structures conversations around moving from a defined goal through honest assessment of current reality, generating options, and committing to a concrete next step. Use it for AI-assisted coaching sessions, personal development prompts, and problem-solving conversations where actionable commitment is the desired output. Always end with a specific, time-bound Will statement to prevent the session from ending in vague intentions. ## What is the GROW Framework? GROW stands for **Goal**, **Reality**, **Options**, and **Will** (also called Way Forward). It is the world's most widely used professional coaching model, developed in the 1980s by Graham Alexander, Alan Fine, and Sir John Whitmore, who popularized it in his landmark book "Coaching for Performance." GROW provides a four-stage structure for moving any conversation or thinking process from aspiration to committed action. - **G — Goal:** Define what you want to achieve - **R — Reality:** Assess the current situation honestly - **O — Options:** Generate a wide range of possibilities - **W — Will:** Commit to a specific plan of action Adapted for AI prompting, GROW transforms the model into a structured thought partner. Instead of asking the AI to simply give advice, a GROW-structured prompt guides the model to work through a disciplined process: clarify the goal with precision, assess the honest starting point without optimism bias, generate a broad range of options without premature filtering, and then commit to a specific, time-bound plan of action. The framework is particularly powerful for AI use because it prevents the two most common failures of advice-seeking conversations: jumping to solutions before the problem is properly understood (skipping Reality), and generating options without ever committing to one (skipping Will). GROW ensures both diagnosis and decision happen before the conversation ends. ## When to Use the GROW Framework 🎯 ### Career Planning Structure conversations about career transitions, promotion goals, or skill development with the AI playing the role of an experienced career coach working through each GROW stage. 🧠 ### Problem Solving Work through complex personal or professional problems by first stating the ideal outcome and honestly assessing where you actually are before generating and evaluating solutions. 🏗️ ### Business Strategy Facilitate strategic planning where a team needs to move from a stated objective through honest current-state assessment to a committed action plan with owners and timelines. ⚖️ ### Decision Making Use GROW to ensure major decisions are made with full awareness of the goal, an honest reading of current constraints, a complete menu of options, and a committed choice with rationale. 💪 ### Goal Setting Transform vague aspirations into specific, measurable goals through the Goal and Reality stages, then build realistic and accountable action plans through Options and Will. 👥 ### Team Facilitation Run team retrospectives, project kick-offs, and planning workshops using GROW as the structural backbone that ensures discussion moves from situation assessment to committed next steps. ## How to Use the GROW Framework 1. 1 ### Goal — Define what you want to achieve State the desired outcome as specifically as possible. A strong Goal is specific, time-bound, and expressed as a positive outcome rather than an escape from a negative situation. "To be leading a team of five engineers by the end of Q3 next year" is a strong Goal. "To stop being a solo contributor" is not. When prompting AI, share your goal clearly and ask the model to help you refine it if it seems vague or ambiguous. 2. 2 ### Reality — Assess the current situation honestly Describe where you actually are right now — not where you wish you were. Include what you have tried, what has worked, what has not, and what specific obstacles exist. The Reality stage requires honesty rather than optimism. Instruct the AI to ask probing questions if your reality assessment seems too abstract, too self-congratulatory, or skips over obvious obstacles. Concrete Reality produces better Options. 3. 3 ### Options — Generate a wide range of possibilities Explore a broad range of options before evaluating any of them. The instinct to jump to the most obvious solution kills creative problem-solving. Ask the AI to generate at least five to eight distinct options, including unconventional ones, before filtering begins. Defer judgment during this stage — quantity and diversity of options matter more than immediate feasibility. Some of the best paths forward emerge from combinations of initially unattractive options. 4. 4 ### Will — Commit to a specific plan of action Select from the options and commit to a concrete plan with named actions, timelines, and accountability mechanisms. The Will stage is where exploration becomes commitment. A weak Will says "I'll look into option three." A strong Will says "By Friday I will email two contacts, schedule one informational interview, and complete one online assessment — and I will report progress to my mentor on Monday." Ask the AI to make the plan specific enough to hold you accountable. ## Prompt Examples Career Coaching — Transitioning from Data Analyst to Product Manager ``` Act as an experienced executive coach specializing in career transitions. Use the GROW coaching model to guide me through the following challenge. Goal: I want to move from my current role as a data analyst into a product management role at a tech company within 12 months. Reality: I have 4 years of experience as a data analyst at a financial services firm. I hold an MBA. I have never held a formal PM role, but I informally led two internal tool projects. I have applied for 3 PM roles in the last 6 months and received no interviews. I am technically strong but unsure how to position my experience for PM roles. For the Options stage: Generate at least 6 concrete, distinct paths I could take to make this transition — from most conventional to most unconventional. For each option, note one key advantage and one significant risk. For the Will stage: Based on the options, help me select the two with the strongest risk-reward profile and build a specific 90-day action plan with weekly milestones and a named accountability mechanism. ``` Business Strategy — Revenue Growth for a Solo Consultancy ``` You are a strategic business advisor using the GROW model to help a small business owner work through a revenue challenge. Goal: To stabilize and grow monthly revenue to 15,000 GBP within 6 months, enabling the hire of a second full-time employee. Reality: The business is a 3-year-old solo UX design consultancy currently generating 9,500 GBP per month. 80% of revenue comes from 2 long-term retainer clients. Business development has been neglected due to full utilization. One retainer client has hinted their budget may be cut in Q4. Current savings runway is 8,000 GBP. Strong referral reputation but no public profile. Options: Generate 7 specific options for closing the revenue gap and reducing client concentration risk. Include at least 2 options that could show measurable results within 60 days. For each option, estimate the effort level (Low / Medium / High) and likely time to first revenue impact. Will: Produce a prioritized Way Forward ranking the top 3 options by speed-of-impact versus effort. Create a 6-week execution plan for the highest-ranked option with named actions and checkpoints. ``` Team Leadership — Building Psychological Safety ``` You are a leadership development coach using the GROW framework. A team leader has come to you with the following situation. Work through each GROW stage in sequence, asking clarifying questions where needed before generating options. Goal: To build a stronger culture of psychological safety in her team of 8 engineers so that people feel comfortable raising concerns and sharing ideas in team meetings. Reality: Currently, 70% of meeting participation comes from 3 out of 8 team members. Post-mortems regularly avoid attribution of mistakes. One-to-one conversations are candid but group discussions are not. The leader has been in the role for 4 months and is still building trust. Generate the Options and Will stages. For Options, provide 5 evidence-based interventions specifically shown to improve psychological safety in engineering teams. For Will, help the leader create a 30-day micro-experiment with clear success metrics she can use to assess progress. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Prevents premature solution-jumping by grounding in honest reality assessment | The Reality stage requires genuine honesty — sugar-coating it produces poor Options | | The Will stage ensures conversations end in commitment rather than exploration | Less effective for purely creative or analytical tasks without a coaching dimension | | Universally understood — widely used in coaching, leadership, and therapy contexts | The Will stage can be vague if the AI is not explicitly prompted for specificity | | Works for individuals, teams, and organizations at any scale | More effective for real goals than hypothetical or academic exercises | | AI can generate richer Options than most people would brainstorm alone | | ## GROW Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Act as an experienced coach using the GROW model. Goal: {{ goal }} Reality: {{ current_situation }} Options: Help me generate at least {{ number_of_options }} options, including ones I have not considered. Will: Help me commit to one concrete next step with a date and a way to measure progress. ``` ## Prompt Flows That Use GROW Framework GROW Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The UX Strategist](https://www.promptedit.app/prompt-flow/the-ux-strategist) · Product Design ## Frequently Asked Questions What does GROW stand for in the context of AI prompting? GROW stands for Goal, Reality, Options, and Will (also called Way Forward). Adapted from the executive coaching world, GROW structures an AI conversation or content piece so it moves through four distinct stages: clarifying what you want to achieve, honestly assessing the current situation, exploring all possible paths forward, and committing to a specific action plan. Who created the GROW model? The GROW model was developed in the 1980s by Graham Alexander, Alan Fine, and Sir John Whitmore, who popularized it in his landmark 1992 book Coaching for Performance. It became the dominant framework in executive coaching globally and has since been adapted for leadership development, career counseling, team facilitation, and AI prompting. How does GROW translate into an AI prompt? You can use GROW in two ways. First, provide all four components as structured input and ask the AI to generate the Options and Will stages — this works well when you know your Goal and Reality but need help with solutions and planning. Second, ask the AI to play a coach role and work through each GROW stage interactively, asking you clarifying questions before generating options. What is the difference between Options and Will in GROW? Options is divergent — it explores the full range of possible paths without commitment, encouraging creativity and breadth. Will (or Way Forward) is convergent — it selects the most promising option and commits to specific actions, timelines, and accountability mechanisms. The shift from Options to Will is one of the most important transitions in the GROW model: from exploring to committing. Can GROW be used for team or organizational decision-making? Yes. GROW scales from individual coaching to team facilitation and organizational strategy. For team use, the Goal becomes a shared objective, Reality describes the team's current capability and constraints, Options generates strategic alternatives, and Will defines the agreed plan with owners and deadlines. Many leadership teams use GROW as their default strategic planning structure. Is GROW suitable for using AI as a personal coach? GROW is the ideal structure for AI coaching conversations. Ask the model to play an executive coach and work through the GROW stages with you: help you articulate your goal clearly, honestly reflect your current reality, brainstorm options you may not have considered, and co-create an action plan with specific commitments. The structured stages keep the conversation productive rather than circular. What should I do if I am stuck in the Reality stage? Being stuck in Reality usually means the current situation is being assessed at too high a level of abstraction. Ask the AI to probe with more specific questions: What has already been tried? What is the specific obstacle? What resources already exist? What data would change your view of the situation? Concreteness in the Reality stage unlocks higher-quality Options in the next stage. ## Origin & Sources 1. Wikipedia (2026). [GROW model](https://en.wikipedia.org/wiki/GROW_model). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/guide [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › GUIDE Framework Structured Framework Intermediate # GUIDE Framework Turn any AI model into a structured advisor — GUIDE closes the loop with an Evaluation component that applies your success criteria to the model's own output. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Goal · Understanding · Information · Direction · Evaluation Best for Teaching, consulting, advisory tasks Unique feature Evaluation component closes the loop **TL;DR:** GUIDE (Goal, Understanding, Information, Direction, Evaluation) is an advisory framework that includes an Evaluation component asking the AI to assess its own output against the stated goal — built-in quality control in a single prompt. Use it for tutoring, expert consulting, and complex advisory tasks where you want the AI to both advise and critique its own advice. Specify in the Evaluation component what success looks like so the self-assessment has a concrete standard to apply. ## What is the GUIDE Framework? GUIDE is a structured prompt engineering framework built around the metaphor of consulting an expert advisor. The acronym stands for **Goal, Understanding, Information, Direction, and Evaluation** — five components that together produce prompts optimized for teaching, consulting, and complex advisory tasks. - **G — Goal:** Define what you want to achieve - **U — Understanding:** Calibrate the model's knowledge baseline - **I — Information:** Supply the raw material - **D — Direction:** Specify the approach and structure - **E — Evaluation:** Set the success criterion What sets GUIDE apart from other structured frameworks is its closing component: Evaluation. Instead of ending when the model delivers an answer, GUIDE asks the model to measure that answer against a defined success criterion. This self-evaluation loop dramatically improves output quality by making the standard of success explicit from the start. GUIDE is particularly powerful when you are not just looking for information but for guidance — when you want the model to think with you, not just respond to you. Its Direction component also gives you precise control over the model's approach, preventing it from defaulting to a generic answer structure when a specific one would serve you better. ## When to Use the GUIDE Framework 🎓 ### Teaching & Tutoring Set a learning goal, share the student's current understanding, provide source material, and define what mastery looks like — the model becomes a structured tutor. 🤝 ### Consulting & Advisory Brief the model like an external consultant: your goal, the context, the data you have, and how you need the advice structured. 📝 ### Technical Documentation Supply raw technical information and direct the model to produce documentation for a specific audience, evaluated against readability criteria. 🏢 ### Business Planning Define a business goal, share market information, and direct the model to build a plan — with Evaluation ensuring the output is actionable and complete. 🔧 ### Performance Optimization Describe a performance problem, share diagnostics, and direct the model to prioritize solutions — evaluated by whether a developer can implement them in a given timeframe. 📣 ### Hiring & Recruitment Write job descriptions, craft interview questions, or evaluate candidate profiles against a defined hiring goal with explicit success criteria. ## How to Use the GUIDE Framework 1. 1 ### Goal — Define what you want to achieve State the end result you are looking for in concrete terms. Be specific about the outcome, not just the activity. "Help me write a job description" is weaker than "Help me write a job description that attracts senior PMs and filters out junior candidates." 2. 2 ### Understanding — Calibrate the model's knowledge baseline Share your relevant experience, expertise level, and any prior attempts. Also signal what the model can assume and what it should not. This prevents both over-explanation and under-explanation in the response. 3. 3 ### Information — Supply the raw material Provide everything the model needs to work with: data, documents, code, specifications, constraints. The richer and more accurate this section, the more grounded and actionable the model's guidance will be. 4. 4 ### Direction — Specify the approach and structure Tell the model how to tackle the problem, not just what to produce. Define the sequence, the angle, the format, and any methodological preferences. This is where you prevent generic, formulaic responses. 5. 5 ### Evaluation — Set the success criterion Describe what a successful response looks like from your perspective. Frame it as a test: "A good answer will allow X person to do Y without needing Z." The model uses this to self-assess and calibrate its output before presenting it. ## Prompt Examples HR & Recruitment — Senior PM Job Description ``` Goal: Help me write a compelling job description for a Senior Product Manager role at a Series B fintech startup that attracts experienced candidates without over-filtering. Understanding: I am a non-technical founder with limited HR experience. I know what the role needs to do functionally, but I struggle to write copy that resonates with strong PM candidates. The company values autonomy, data-driven decision making, and cross-functional collaboration. Information: The role owns the core payments product. The team is 4 engineers and 1 designer. The PM will report directly to me. Salary range is $160k–$185k with meaningful equity. We are fully remote with quarterly in-person sprints. Direction: First, identify what top PM candidates look for in a job description. Then draft a job description that leads with the mission and opportunity, follows with responsibilities and qualifications, and ends with a clear benefits and culture section. Avoid jargon and buzzwords. Evaluation: The result is successful if a PM with 5+ years of experience would read the description and immediately understand the scope of ownership, the growth opportunity, and whether they are a fit — without needing to ask clarifying questions. ``` Engineering — AWS Lambda Cold-Start Optimization ``` Goal: Advise me on how to reduce the cold-start latency of a Python AWS Lambda function from ~3 seconds to under 800ms. Understanding: I am a backend developer with 3 years of Python experience and 1 year of AWS Lambda experience. I understand the concept of cold starts but have not profiled my function to know which layer is the slowest. The function uses SQLAlchemy, Pandas, and a custom internal library. Information: The Lambda is configured with 512MB RAM, Python 3.11 runtime, deployed as a zip package (current size: 48MB). It connects to an RDS PostgreSQL instance via a connection pool. Cold starts happen for approximately 15% of invocations in production. Direction: Start by identifying the most common root causes of cold-start latency for Python Lambdas of this type. Then provide a prioritized list of optimizations, starting with the highest-impact, lowest-effort changes. Include specific configuration values or code patterns where applicable. Evaluation: A good answer will give me a clear action plan I can implement in one sprint (5 business days), with each recommendation including an expected latency improvement in milliseconds or a percentage range. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Evaluation component makes success criteria explicit and improves output quality | Five components require meaningful upfront investment per prompt | | Direction component prevents generic, formulaic responses | Evaluation component is only as useful as the success criteria you write | | Scales well from simple advisory tasks to complex consulting scenarios | Less suited for quick, factual queries where zero-shot is more efficient | | Intuitive consultation metaphor accessible to non-technical users | | ## GUIDE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Goal: {{ goal }} Understanding: I already know {{ my_knowledge_level }}. Information: {{ source_material }} Direction: {{ approach_and_structure }} Evaluation: A good answer {{ success_criterion }}. ``` ## Frequently Asked Questions What does GUIDE stand for in prompt engineering? GUIDE stands for Goal, Understanding, Information, Direction, and Evaluation. It is a five-component framework designed for advisory, teaching, and consulting-style prompts where you want the model to not only answer a question but genuinely guide you toward a well-reasoned outcome. What makes GUIDE different from other structured frameworks? GUIDE's defining feature is its Evaluation component, which asks the model to assess whether its own output meets the goal. Most frameworks end at the answer — GUIDE closes the loop by including a success criterion that the model applies to its own response. This makes it especially effective for complex or high-stakes tasks. What should I write in the 'Direction' component? Direction tells the model how to approach the task — the method, the sequence of steps, or the angle of attack. For example: 'Start by identifying the root cause, then propose three solutions ranked by cost-effectiveness.' Direction prevents the model from choosing an approach that is valid but not useful to you. How is 'Understanding' used in GUIDE vs. in QUEST? In QUEST, Understanding is about providing background context from your perspective. In GUIDE, Understanding also includes what the model's starting knowledge base should be — it calibrates both the context you bring and the knowledge the model should assume or defer on. This makes GUIDE more consultation-focused. Can I use GUIDE for technical documentation tasks? Yes. GUIDE works well for documentation when you use the Goal to specify the audience, Information to supply the raw technical details, Direction to set the structure, and Evaluation to define what 'good' documentation looks like (e.g., 'a junior developer should understand every step without prior context'). What goes in the 'Information' component? Information is the data, facts, or materials the model needs to do the task. This is where you paste in raw data, provide reference documents, share URLs, include code snippets, or supply any other input that the model should process or reason from. It is the raw material for the model's work. Is GUIDE suitable for non-technical users? Absolutely. GUIDE is particularly well-suited for non-technical users who need expert-level guidance. The framework's consultation metaphor — briefing an expert advisor — is intuitive. You describe what you want (Goal), what you know (Understanding), what you have (Information), how you want it approached (Direction), and what success looks like (Evaluation). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/idea [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › IDEA Framework Structured Framework Beginner # IDEA Framework The creative prompt framework built for iteration — IDEA's Adjustments component encodes your feedback rules upfront so every revision request produces exactly the right change. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Intent · Details · Examples · Adjustments Best for Creative tasks, iterative work, design briefs Unique feature Adjustments component enables in-prompt iteration **TL;DR:** IDEA (Intent, Details, Examples, Adjustments) is an iterative creative framework that builds revision guidance directly into the initial prompt, reducing back-and-forth on creative and design tasks. Use it for brand content, UX copy, and any creative deliverable that typically requires 2-3 revision rounds. Include the Adjustments component to pre-answer the most likely revision requests before the AI even produces its first draft. ## What is the IDEA Framework? IDEA is a structured prompt engineering framework designed around the reality that creative and design tasks are inherently iterative. The acronym stands for **Intent, Details, Examples, and Adjustments** — four components that together create a prompt optimized not just for a good first output but for efficient collaboration across multiple revision cycles. - **I — Intent:** State what you want to accomplish - **D — Details:** Specify all concrete requirements - **E — Examples:** Provide style and tone references - **A — Adjustments:** Encode your revision rules The framework's most distinctive innovation is the **Adjustments** component. Instead of waiting until the model produces an output to explain what you want changed, IDEA asks you to encode your revision logic upfront. When you tell the model "if I say 'punchier', do X" before the first output, every subsequent revision request is processed with complete clarity — the model knows exactly what each feedback signal means. This approach is particularly valuable for creative tasks where the definition of "better" is personal and difficult to articulate in the moment. By thinking through your likely adjustments before you see the first draft, you also clarify your own preferences — often producing a better first output even before any iteration occurs. ## When to Use the IDEA Framework ✍️ ### Copywriting & Content Email campaigns, landing pages, social media posts, and headlines — any copy that will go through multiple rounds of revision before publication. 🎨 ### Design Briefs Brief a model to generate UI copy, naming options, taglines, or brand messaging where style and tone preferences are difficult to articulate without iteration. 📖 ### Creative Writing Short stories, blog posts, scripts, and other long-form creative content where the first draft establishes direction and subsequent drafts refine it. 🖥️ ### UI & Product Copy Microcopy, onboarding flows, error messages, and button labels — small-scale text where the right wording requires several rounds of refinement. 📣 ### Marketing & Brand Brand voice development, positioning statements, and campaign concepts that need to be tested against multiple angles before settling on the right direction. 💡 ### Ideation & Brainstorming Generate an initial set of ideas, then use Adjustments to steer subsequent rounds toward the most promising directions without starting from scratch. ## How to Use the IDEA Framework 1. 1 ### Intent — State what you want to accomplish Define the purpose of the creative output in one or two sentences. Include who it is for, what it should do, and what context it will appear in. Intent anchors all subsequent components and prevents the model from interpreting Details and Examples in a vacuum. 2. 2 ### Details — Specify all concrete requirements List every constraint and requirement: word count, format, tone, audience, what to include, what to avoid, and any technical or brand guidelines. The Details component is where you prevent the model from making assumptions on things you care about deeply. 3. 3 ### Examples — Provide style and tone references Share one to three examples that illustrate the style, tone, or structure you are aiming for. These can be direct quotes, links, or descriptions of existing work. Explain what aspect of each example you want the model to draw from — style, structure, register, or approach. 4. 4 ### Adjustments — Encode your revision rules Anticipate the feedback you are most likely to give and tell the model what each signal means. Use the pattern: "If I say [feedback word], then [specific action]." This is the component that makes IDEA uniquely powerful for iterative work — it transforms vague feedback into precise instructions the model can act on reliably. ## Prompt Examples Marketing — Product Launch Email for AI Writing Tool ``` Intent: Write a product launch email announcing the release of our new AI writing assistant, "Compose", to our existing customer base of 8,000 small business owners. Details: The email should be 180–220 words. Subject line must create curiosity without being clickbait. The body should lead with the key benefit (save 3 hours per week on business writing), mention 3 core features in a scannable format, include one social proof element (we have a quote from a beta user), and end with a single clear CTA linking to the product page. Tone: warm, confident, and direct. No exclamation marks. Examples: The tone and structure of Basecamp's product announcement emails — plain-spoken, benefit-first, no hype. Reference: "Basecamp 3 is here. Here's what it does for you..." style of opening. Adjustments: If I say "punchier", tighten each sentence and lead with a stronger verb. If I say "more formal", shift to a professional but still approachable register. If I say "different angle", rewrite the opening to lead with the time-saving benefit differently — try a before/after structure instead. ``` Brand Copy — B2B SaaS Landing Page Headlines ``` Intent: Design a landing page headline and three supporting subheadlines for a B2B SaaS product that automates expense reporting for finance teams. Details: The main headline should be under 10 words and communicate the primary value proposition: eliminating manual expense reconciliation. Each subheadline (25–40 words) should support a different pillar: speed, accuracy, and integration with existing accounting software. Audience: Finance Directors and CFOs at companies with 100–500 employees. Register: authoritative, outcome-focused, minimal jargon. Avoid the word "seamless." Examples: Style reference — Stripe's homepage copy: clear, specific, confident without being aggressive. Also reference Rippling's finance product messaging for the target audience register. Adjustments: If I say "more technical", introduce specific metrics (e.g., "reduces close time by 4 days"). If I say "simpler", strip each headline to its core claim with no qualifiers. If I say "try again", start from a completely different angle — consider leading with the problem rather than the solution. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Adjustments component eliminates ambiguous feedback and accelerates iteration | Requires upfront thinking about your revision preferences before seeing the first output | | Examples component communicates style more precisely than description alone | Less suited for analytical or research tasks where iteration is not the primary workflow | | Beginner-friendly — four components are intuitive and easy to apply | Examples component can introduce unintended style bias if references are poorly chosen | | Works across any domain where iteration is expected, not just creative writing | | ## IDEA Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Intent: {{ intent }} Details: {{ requirements }} Examples: Match the style of: {{ reference_examples }} Adjustments: When I ask for revisions, {{ revision_rules }} ``` ## Frequently Asked Questions What does IDEA stand for in prompt engineering? IDEA stands for Intent, Details, Examples, and Adjustments. It is a four-component framework designed for creative tasks, iterative work, and design briefs. Its defining feature is the Adjustments component, which builds a feedback loop directly into the initial prompt — telling the model how to respond when you request changes. What makes IDEA different from other prompt frameworks? Most prompt frameworks are designed to get the right answer on the first try. IDEA is designed for iterative work — where the first output is a starting point, not a final deliverable. The Adjustments component explicitly tells the model how to handle revision requests, making it ideal for creative collaboration where you refine output over multiple turns. What should I include in the 'Adjustments' component? Adjustments should tell the model your revision preferences and how to interpret feedback signals. For example: 'If I say shorter, reduce by at least 30%. If I say more formal, shift the register to match a Harvard Business Review article. If I say try again, produce a completely different approach rather than a slight variation.' The more specific your adjustment rules, the faster you converge on the result you want. Can I use IDEA for non-creative tasks? IDEA works for any task that benefits from iteration — not just creative writing. Code generation, UI copy, email templates, and marketing headlines all benefit from the IDEA structure because they are rarely perfect on the first attempt. The framework's iterative nature is its core advantage, regardless of the domain. How detailed should the 'Details' component be? Details should cover every specific requirement that is not captured by the Intent. Length, tone, format, technical constraints, brand guidelines, audience, and any explicit inclusions or exclusions all belong here. The Details component is where you prevent the model from making assumptions about things you actually have strong preferences about. What is the difference between Examples and Details in IDEA? Details are your requirements — what the output must do or be. Examples are references — things that illustrate the style, tone, or structure you are aiming for, even if the content is completely different. For instance, Details might say 'use a conversational tone', while Examples might link to or quote a specific piece of writing that embodies that tone better than any description could. How many examples should I include in the Examples component? One to three examples is typically sufficient. Too few and the model has insufficient signal about your preferences. Too many and the model may average across conflicting styles rather than applying a clear direction. If your examples vary significantly in style, note which aspect of each example you want the model to take inspiration from. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/ieei [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › IEEI Framework Structured Framework Beginner # IEEI Framework Structure every prompt with four clear layers — Inform, Explain, Example, Input — to give AI models everything they need before touching your content. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Inform, Explain, Example, Input Best for Educational content, how-to guides Token cost Moderate **TL;DR:** IEEI (Inform, Explain, Example, Input) is an educational prompt structure that establishes context and demonstrates quality before presenting the content to process — ensuring the AI responds with calibrated understanding rather than default assumptions. Use it for structured analysis, classification tasks, and educational content generation where the quality bar matters. The Example component is the most important: it shows rather than tells what a good response looks like. ## What is the IEEI Framework? IEEI is a four-part prompt structure built around the sequence Inform, Explain, Example, Input. Each layer serves a distinct purpose: the first two establish context and expectations for the model, the third demonstrates desired output quality, and the fourth delivers the actual content to process. - **I — Inform:** Establish what the AI should know - **E — Explain:** Clarify the task and context in detail - **E — Example:** Provide a reference or sample output - **I — Input:** Provide the actual content to process Unlike freeform prompting, IEEI separates concerns so the model never has to guess what role it should play, what the task requires, or what good output looks like. The result is more consistent, on-target responses — especially for tasks you repeat often or need to hand off to others. The framework is particularly well suited to educational and explanatory content because each of its four steps mirrors how good teachers frame lessons: establish prior knowledge, explain the learning objective, show a worked example, and then practice with new material. ## When to Use the IEEI Framework 📚 ### Educational Content Create lessons, explainers, or concept summaries where consistent depth and pedagogical structure are critical. 📝 ### How-To Guides Write step-by-step instructions for software, processes, or skills where examples ground abstract steps in reality. 🔄 ### Batch Processing Process many similar inputs — product descriptions, article summaries, FAQ answers — with a single reusable prompt template. ✍️ ### Style-Matched Writing Replicate a specific writing style or tone by providing a style sample as the Example section. 🧑‍🏫 ### Knowledge Transfer Translate complex technical documentation into accessible language for different audience levels. 📊 ### Report Generation Produce structured reports from raw data where a reference report in the Example section ensures consistent formatting. ## How to Use the IEEI Framework 1. 1 ### Inform — Establish what the AI should know Open with the model's role or the knowledge context it needs. This is not a vague instruction like "be helpful" — it is a precise domain or persona statement. For example: "You are a senior technical writer who specialises in developer documentation for SaaS APIs." 2. 2 ### Explain — Clarify the task and context in detail Describe what you want the model to produce, for whom, in what format, and under what constraints. Cover tone, length, audience level, and any things to avoid. The more specific this section, the less guesswork the model does. 3. 3 ### Example — Provide a reference or sample output Supply a concrete example of what a good response looks like. This acts as a one-shot demonstration that calibrates vocabulary, depth, structure, and tone without lengthy meta-instructions. The example does not need to be perfect — it just needs to show the right shape. 4. 4 ### Input — Provide the actual content to process Clearly delimit the material the model should work on — use markers like "INPUT:" or triple backticks to avoid confusion with the instructional sections above. Everything before this was setup; this is the work. ## Prompt Examples IEEI — Tutorial Writing ``` INFORM: You are an experienced software educator who writes beginner-friendly Python tutorials for adult learners switching careers into tech. EXPLAIN: Write a short tutorial section (150–200 words) that explains the concept in the INPUT below. Use plain language, avoid jargon unless you define it immediately, include one analogy, and end with a one-sentence takeaway the reader should remember. EXAMPLE: Concept: Variables "Think of a variable as a labelled box. You put a value inside the box, stick a label on the outside, and whenever you need that value later, you just read the label instead of memorising the contents. In Python, creating a variable is as simple as writing: name = "Alice" Here, 'name' is the label and "Alice" is what's inside the box. You can change the contents at any time by assigning a new value. Takeaway: Variables let you store and reuse data without repeating yourself." INPUT: Concept: For loops ``` IEEI — Document Summarisation ``` INFORM: You are a professional business analyst who specialises in condensing lengthy corporate reports into executive-ready summaries. EXPLAIN: Summarise the document provided in the INPUT section. The summary must be exactly three paragraphs: (1) the core finding or recommendation, (2) the key supporting evidence, and (3) the suggested next steps. Write for a C-suite audience with no technical background. Use active voice and keep each paragraph under 80 words. EXAMPLE: Summary of Q2 Market Expansion Report: The Asia-Pacific market presents the strongest growth opportunity for the next fiscal year, with projected revenue upside of 34% above current targets. Three primary factors support this conclusion: rising smartphone penetration in Tier 2 cities, a favourable regulatory shift effective January, and two competitor withdrawals that have left a gap in the mid-market segment. Leadership should authorise a pilot launch in two target cities by Q3, commission a localisation study within 60 days, and appoint a regional lead no later than end of this quarter. INPUT: [Paste your document here] ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Highly beginner-friendly with four intuitive, labelled sections | Longer prompts mean higher token usage per request | | The Example step acts as a built-in quality calibrator | Crafting a good Example section takes time upfront | | Produces consistent results across repeated or batched tasks | Can feel over-structured for simple, one-off tasks | | Works well as a reusable template for content teams | | ## IEEI Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Inform: {{ background_knowledge }} Explain: {{ task_explanation }} Example: {{ reference_example }} Input: {{ content_to_process }} ``` ## Prompt Flows That Use IEEI Framework IEEI Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Onboarding Writer](https://www.promptedit.app/prompt-flow/the-onboarding-writer) · Documentation ## Frequently Asked Questions What is the IEEI framework? IEEI stands for Inform, Explain, Example, Input. It is a four-part prompt structure designed to give the AI all the context it needs before processing your actual content. By informing the model of its role, explaining the task in detail, showing a reference example, and then supplying the input, you dramatically reduce ambiguity and improve output quality. How is IEEI different from a simple instruction prompt? A simple instruction prompt tells the AI what to do. IEEI tells it what to do, why, how good looks like, and then gives it the raw material to work with. The added layers — especially the Example step — act as an implicit quality bar the model tries to match or exceed. When should I use the IEEI framework? IEEI shines when you need consistent, educational, or explanatory content: writing tutorials, summarising documents, teaching concepts, creating how-to guides, or processing large batches of similar inputs. It is especially effective when you want every output to follow a consistent style or depth. Is IEEI suitable for beginners? Yes. IEEI is one of the most beginner-friendly structured frameworks because each of its four steps maps naturally to questions anyone would ask: 'What role should the AI play?', 'What exactly do I want?', 'What does a good answer look like?', and 'What is my actual content?' Following the four labels keeps prompts organised without requiring advanced prompt-engineering knowledge. Can I skip the Example step in IEEI? You can, but the Example step is where much of IEEI's power comes from. It functions as a one-shot demonstration that calibrates tone, format, depth, and vocabulary. Skipping it turns IEEI into a two-element structure closer to plain role + task prompting, which is less reliable for complex or style-sensitive outputs. How long should each IEEI section be? Inform: one or two sentences establishing the AI's role or knowledge domain. Explain: a short paragraph covering the task, constraints, format, and audience. Example: a concrete sample that demonstrates the desired output — ideally 50–200 words. Input: the actual content to process. Longer is not always better; clarity in each section matters more than length. What models work best with the IEEI framework? IEEI works well with any instruction-following model, including GPT, Claude, and Gemini. Because it front-loads context before the input, it is particularly effective for models that attend strongly to system-level and early-message context. It also works reliably in API settings where you split the four parts across system and user messages. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/least-to-most [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Least-to-Most Prompting Reasoning Intermediate # Least-to-Most Prompting Break a hard problem into simpler sub-problems, then solve them from easiest to hardest, feeding each answer into the next. Planning and solving become two separate, checkable stages. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Zhou et al., Google (2022) Stages Decompose → solve sequentially Best for Problems harder than your examples Token cost Medium **TL;DR:** Least-to-Most Prompting (Zhou et al., 2022) first asks the model to split a problem into simpler sub-problems, then solves them in order with each answer available to the next step. It generalises to problems harder than the examples in the prompt, reaching 99% on the SCAN benchmark where Chain-of-Thought reached 16%. Use it for multi-step problems where the order of work matters. ## What is Least-to-Most Prompting? Least-to-Most Prompting comes from the paper *Least-to-Most Prompting Enables Complex Reasoning in Large Language Models* by Denny Zhou and colleagues at Google, published at ICLR 2023. It was designed to fix a specific weakness of Chain-of-Thought: models do well on problems similar to the examples in the prompt, but struggle when the real problem is harder than anything they were shown. The technique borrows a strategy from education research: teach by moving from the least difficult part to the most difficult. It has two stages: - **Decomposition:** the model lists the sub-problems that need to be solved, ordered so that each builds on the previous ones. - **Sequential solving:** the model solves each sub-problem in turn, with the question and all previous answers in context, until the final sub-problem answers the original question. On SCAN, a benchmark that tests whether models can combine known pieces into new instructions, Least-to-Most with 14 examples reached at least 99% accuracy, while Chain-of-Thought reached about 16%. ## When to Use Least-to-Most Prompting 🧮 ### Multi-Step Math Word problems where intermediate quantities must be computed before the final one can be. 💻 ### Coding Tasks Features that need a data model, then an API, then UI changes, each depending on the previous step. 🗂️ ### Project Planning Turn a large goal into ordered milestones, then detail each milestone with the previous ones as context. 🔍 ### Research Questions Answer background questions first, then the comparative or causal question that depends on them. 🧩 ### Compositional Instructions Commands built from smaller commands, such as data transformations or query building. 📊 ### Data Analysis Clean, aggregate, then interpret: each stage needs the output of the one before. ## How to Use Least-to-Most Prompting 1. 1 ### Ask for a decomposition "To answer this, what simpler questions do we need to answer first?" Ask for a numbered list ordered from simplest to the final question. 2. 2 ### Review the sub-problems Check that answering all of them really answers the original question. For important work, fix the list before solving. 3. 3 ### Solve one sub-problem at a time Give the model the original question, all previous sub-problems with their answers, and the next sub-problem. 4. 4 ### Answer the original question last The final sub-problem should be the original question itself, now answered with everything solved so far in context. ## Prompt Examples Single-Prompt Least-to-Most ``` Problem: Elsa has 5 apples. Anna has 2 more apples than Elsa. Kristoff has twice as many as Anna. How many apples do they have together? Stage 1: List the sub-questions we must answer first, from simplest to the original question. Stage 2: Answer each sub-question in order. Before each one, restate the answers you already have. Final answer on the last line. ``` Feature Planning (two calls) ``` Call 1: We want to add "export invoices as PDF" to our billing web app (Node.js API, React frontend, Postgres). Decompose this into the smallest set of sub-problems we must solve, ordered so that each depends only on earlier ones. Return a numbered list only. Call 2 (repeat per sub-problem): Goal: add "export invoices as PDF". Already solved: {{ previous_sub_problems_and_answers }} Now solve sub-problem {{ n }}: {{ sub_problem }} Be concrete: files, functions, data structures. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Generalises to problems harder than the examples shown | More calls or longer output than single-pass reasoning | | Decomposition is visible and can be reviewed before solving | A wrong decomposition derails every later step | | Each step has a narrow focus, reducing errors | Overkill for simple, single-step tasks | | Maps directly onto multi-call pipelines and agents | Context grows with each solved sub-problem | ## Least-to-Most Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Problem: {{ problem }} Stage 1: Break the problem into the simpler sub-problems we must solve first. Number them from simplest to the original problem. Stage 2: Solve the sub-problems in order. Before each one, restate the answers you already have. Final answer: {{ output_format }} ``` ## Frequently Asked Questions What is Least-to-Most Prompting? Least-to-Most Prompting is a two-stage technique introduced by Denny Zhou and colleagues at Google (2022, ICLR 2023). First the model decomposes a complex problem into a list of simpler sub-problems. Then it solves them in order, from easiest to hardest, and each answer is added to the context for the next sub-problem. How is Least-to-Most different from Chain-of-Thought? Chain-of-Thought reasons through the whole problem in one pass. Least-to-Most separates planning from solving: it first makes the decomposition explicit, then solves each sub-problem with the previous answers available. That makes it better at problems harder than the examples shown in the prompt, which is where Chain-of-Thought tends to break down. What results did Least-to-Most achieve? On the SCAN compositional generalisation benchmark, GPT-3 (code-davinci-002) with Least-to-Most reached at least 99% accuracy using just 14 examples, compared with about 16% using Chain-of-Thought. Specialised neural-symbolic models had needed over 15,000 training examples for similar results. Is Least-to-Most the same as Prompt Chaining? They are related. Prompt Chaining is the general pattern of feeding one prompt's output into the next, with steps you design by hand. Least-to-Most is a specific reasoning method where the model itself produces the decomposition and the sub-problems are ordered from simplest to hardest. Can I run Least-to-Most in a single prompt? Yes. Ask for a numbered list of sub-problems first, then instruct the model to solve them one at a time, restating the previous answers before each step. For long or high-stakes tasks, separate calls give you more control and let you check the decomposition before solving. When should I avoid Least-to-Most? For simple tasks the decomposition step is overhead. It also fails if the decomposition is wrong, so review the sub-problem list for complex work, or ask the model to check that solving all sub-problems actually answers the original question. ## Origin & Sources 1. Zhou et al. (2022). [Least-to-Most Prompting Enables Complex Reasoning in Large Language Models](https://arxiv.org/abs/2205.10625) . arXiv / ICLR 2023. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/multi-agent-debate [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Multi-Agent Debate Advanced Technique Advanced # Multi-Agent Debate (MAD) Deploy multiple AI agents as structured debaters — each arguing a distinct position — then synthesize their arguments into answers that have survived adversarial scrutiny from every angle. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Research basis Du et al. (2023), Liang et al. (2023) Typical setup 3–5 agents, 2–4 debate rounds + judge Key benefit Surfaces weaknesses single-agent responses miss Best for Fact verification, decisions, complex analysis **TL;DR:** Multi-Agent Debate (MAD) runs multiple AI agents with distinct perspectives through adversarial rounds, then uses a judge agent to synthesize the debate into a final answer — significantly improving factual accuracy over single-agent approaches. Use it for high-stakes factual questions, policy analysis, and complex decisions where unchallenged reasoning leads to overconfident errors. Assign agents opposing stances, run 2-3 debate rounds, then have a neutral judge synthesize the strongest arguments. ## What is Multi-Agent Debate? Multi-Agent Debate (MAD) is a prompting architecture in which multiple AI agent instances — each given a distinct role, perspective, or epistemic stance — independently analyze a problem, then engage in structured adversarial debate across multiple rounds before a judge agent synthesizes a final answer. The technique was formalized in research by Du et al. (2023) "Improving Factuality and Reasoning in Language Models through Multiagent Debate" and Liang et al. (2023), which demonstrated that agents that debate achieve significantly higher accuracy on factual reasoning tasks compared to single-agent approaches — including those using Chain-of-Thought or Self-Consistency. The core mechanism exploits a well-known property of LLMs: they tend to maintain and defend a stated position better when they receive explicit counter-arguments to respond to. By forcing agents into adversarial exchange, MAD creates a dynamic where: - **Weaknesses get surfaced:** The Skeptic agent will attack the Proponent's logical gaps, factual errors, and unexamined assumptions — things a single agent generating a complete answer would paper over. - **Sycophancy is reduced:** Agents arguing assigned positions cannot simply agree with each other — they must defend or revise based on evidence, not social pressure. - **Synthesis is richer:** The Judge agent produces a conclusion that has been pressure-tested from multiple angles, with clear identification of what is certain versus contested. ## When to Use Multi-Agent Debate 🔬 ### Scientific & Medical Fact Verification Claims in medicine, nutrition, and science are often contested or nuanced. MAD with a Proponent, Skeptic, and Evidence Quality Analyst produces more calibrated assessments than a single-agent summary that may over- or under-state evidence strength. ⚖️ ### Legal & Ethical Analysis Legal questions rarely have unambiguous answers. Assigning Plaintiff, Defense, and Judge roles models the adversarial legal reasoning process and surfaces the strongest arguments on each side before reaching a conclusion. 📈 ### Strategic Business Decisions High-stakes decisions about market entry, acquisitions, product strategy, or resource allocation benefit from explicit best-case (Optimist), worst-case (Pessimist), and contrarian (Devil's Advocate) agents before synthesis. 🛡️ ### Risk & Failure Mode Analysis Engineering, security, and operational risk assessments require someone to actively try to break proposed solutions. A dedicated Red Team agent arguing against every proposed safeguard surfaces vulnerabilities that a cooperative single-agent analysis misses. 🗳️ ### Policy & Governance Analysis Policy proposals have stakeholders with legitimate competing interests. Multi-agent debate with Stakeholder A, Stakeholder B, and a Policy Analyst models the real-world negotiation and produces recommendations that have grappled with trade-offs explicitly. 🎓 ### Educational Socratic Dialogue MAD can simulate Socratic teaching: one agent argues a common misconception, another refutes it with correct reasoning, and the exchange models the learning process more effectively than a single-agent explanation. ## How to Implement Multi-Agent Debate 1. 1 ### Define the debate question and agent roles Formulate a precise, debatable question — vague questions produce vague debates. Then assign 3–5 agent roles that represent genuinely distinct epistemic positions relevant to the question. For factual claims: Proponent, Skeptic, Evidence Analyst. For decisions: Optimist, Pessimist, Devil's Advocate. For multi-domain analysis: domain specialist roles (financial, technical, legal). Each role must have a meaningful perspective that could legitimately lead to a different conclusion — otherwise the debate is performative rather than substantive. 2. 2 ### Run Round 1 — Independent opening arguments Prompt each agent independently in separate LLM calls, providing only the question and their role. Do not let agents see each other's Round 1 responses yet — independence at this stage ensures the opening arguments reflect genuine perspective differences rather than anchoring to a first-mover's framing. Collect all Round 1 responses before proceeding. 3. 3 ### Run Round 2 — Cross-examination and rebuttal Now share all Round 1 responses with each agent and prompt them to respond: acknowledge valid points from other agents, challenge weak arguments, and refine or defend their own position. This is where the debate becomes genuinely productive — agents must engage with opposing arguments rather than simply restating their opening. Run 2–3 rebuttal rounds depending on task complexity and budget. 4. 4 ### Run the Judge synthesis Provide the complete debate transcript to a Judge agent and instruct it to synthesize a final verdict. The Judge should not simply average positions — instruct it to identify the strongest arguments, name remaining uncertainties, state its confidence level, and give conditions under which the conclusion would change. A high-quality synthesis is the final deliverable; invest in the Judge prompt carefully. ## Prompt Examples Round 1 — Debate Setup and Opening Arguments ``` ## Multi-Agent Debate — Fact Verification Setup DEBATE TOPIC: "Intermittent fasting produces greater long-term weight loss than continuous caloric restriction." AGENT ROLES: - Agent 1 (Proponent): Argues FOR the claim with supporting evidence - Agent 2 (Skeptic): Challenges the claim, identifies methodological weaknesses - Agent 3 (Neutral Analyst): Evaluates the evidence standards and study quality ROUND 1 INSTRUCTIONS (run separately for each agent): --- Agent 1 Prompt --- You are a nutrition researcher arguing FOR the following claim. Draw on the strongest available evidence. Present 3 key supporting arguments with citations or references to study types. Be specific and rigorous. Claim: "Intermittent fasting produces greater long-term weight loss than continuous caloric restriction." Your opening argument (max 200 words): --- Agent 2 Prompt --- You are a critical nutrition scientist who is SKEPTICAL of the following claim. Your role is to challenge the evidence, identify methodological limitations, and present counter-evidence. Be rigorous, not dismissive. Claim: [same claim] Your opening challenge (max 200 words): --- Agent 3 Prompt --- You are a systematic review expert evaluating evidence quality. Assess both positions based on: study design quality, sample sizes, follow-up duration, and publication bias risk. [Agent 1 argument]: [PASTE] [Agent 2 argument]: [PASTE] Your evidence quality assessment (max 150 words): ``` Round 2 — Rebuttal Phase ``` ## Multi-Agent Debate — Round 2 Rebuttal DEBATE CONTEXT (Round 1 complete): Agent 1 (Proponent) argued: [PASTE ROUND 1 AGENT 1 RESPONSE] Agent 2 (Skeptic) argued: [PASTE ROUND 1 AGENT 2 RESPONSE] Agent 3 (Analyst) assessed: [PASTE ROUND 1 AGENT 3 RESPONSE] --- ROUND 2 INSTRUCTIONS — Rebuttal Phase Agent 1 Rebuttal Prompt: You are the Proponent from Round 1. You have now read the Skeptic's challenge and the Analyst's assessment. Address the strongest objections raised. Which of your Round 1 arguments do you maintain? Which do you modify? Are there any concessions you must make based on the evidence quality critique? (max 150 words) Agent 2 Rebuttal Prompt: You are the Skeptic from Round 1. You have read the Proponent's argument and the Analyst's assessment. Which of your challenges were validated? Which require revision? What is the strongest remaining objection? (max 150 words) ``` Judge Synthesis — Final Verdict ``` ## Multi-Agent Debate — Judge Synthesis Prompt FULL DEBATE TRANSCRIPT: Round 1: Proponent: [PASTE] Skeptic: [PASTE] Analyst: [PASTE] Round 2: Proponent Rebuttal: [PASTE] Skeptic Rebuttal: [PASTE] --- JUDGE SYNTHESIS INSTRUCTIONS: You are an impartial judge who has observed the full debate above. Your task is to produce a final, evidence-based verdict. Your synthesis must: 1. State the final verdict clearly (supported / not supported / partially supported) 2. Identify the 2–3 strongest arguments from either side that most influenced the verdict 3. Name the key uncertainty or evidence gap that prevents a definitive conclusion 4. Give a confidence level (low / medium / high) with brief justification 5. State one specific condition under which the verdict would change Do not simply average the positions. Make a judgment call and defend it. Maximum 300 words. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Significantly improves factual accuracy and reasoning quality on hard tasks | High token cost — multiple agents across multiple rounds multiplies LLM calls | | Surfaces weaknesses and counter-arguments that single-agent prompting misses | High latency — sequential debate rounds cannot be parallelized | | Reduces sycophancy — agents must defend positions under adversarial pressure | Requires careful role design — poorly assigned roles produce superficial debate | | Produces more calibrated, nuanced conclusions with acknowledged uncertainty | Overkill for straightforward tasks where a single well-structured prompt suffices | ## Multi-Agent Debate Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Question: {{ question }} Simulate a debate between {{ number_of_agents }} experts: {{ agent_perspectives }} Round 1: each expert gives an independent answer with reasoning. Round 2: each expert reads the others' answers and critiques or updates their own position. Round 3: a neutral judge summarises the strongest arguments, notes unresolved disagreements, and gives the final answer with a confidence level. ``` ## Prompt Flows That Use Multi-Agent Debate Multi-Agent Debate is one step in these 2 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The API Designer](https://www.promptedit.app/prompt-flow/the-api-designer) · Software Development - [The Debate Judge](https://www.promptedit.app/prompt-flow/the-debate-judge) · Decision-Making ## Frequently Asked Questions What is Multi-Agent Debate (MAD)? Multi-Agent Debate (MAD) is a prompting strategy where multiple AI agent instances — each assigned a distinct perspective or role — independently analyze a problem, then argue their positions against one another in structured rounds, before a final synthesis or judge agent produces a consolidated answer. The debate process forces explicit consideration of opposing viewpoints and surfaces assumptions that a single-agent response would leave unexamined. How does Multi-Agent Debate improve answer quality? MAD improves quality through several mechanisms: it prevents premature closure on the first plausible answer, forces each position to be defended under adversarial scrutiny (which surfaces logical weaknesses), incorporates diverse framing of the same problem, and reduces the sycophancy bias where a single model tends to agree with whatever the user implies. Research (Du et al., 2023; Liang et al., 2023) shows that multi-agent debate consistently outperforms single-agent prompting on reasoning, fact-checking, and complex decision tasks. Is Multi-Agent Debate the same as Self-Consistency? No. Self-Consistency samples multiple independent answers from the same prompt and takes a majority vote — there is no interaction between agents. Multi-Agent Debate involves actual argumentative exchange: agents see each other's positions and respond to them, building on or refuting prior arguments. The interaction is the key differentiator. MAD produces richer, more nuanced synthesis because agents must reconcile conflicting positions rather than simply aggregating them. How many agents and debate rounds are optimal? Research and practice suggest 3–5 agents per debate and 2–4 rounds of exchange. Below 3 agents, the debate lacks genuine diversity of perspective. Above 5, the synthesis becomes unwieldy and the incremental benefit of additional agents diminishes. For rounds, 2–3 is sufficient for most tasks — agents typically converge or have clarified their disagreements by round 3. More rounds add cost without proportional quality improvement. What roles should the debate agents have? Agent roles depend on the task type. For factual verification: Proponent (argues the claim is true), Skeptic (challenges the claim), and Domain Expert (evaluates evidence rigor). For decisions: Optimist (best-case), Pessimist (worst-case/risks), and Devil's Advocate (challenges any consensus). For analysis: multiple domain-specific experts (financial analyst, legal analyst, technical analyst) whose perspectives must all be reconciled. The Judge agent synthesizes all positions into a final answer. Can Multi-Agent Debate be run with a single LLM API? Yes. You do not need multiple model instances or a multi-agent framework. MAD can be simulated with a single LLM by providing the full debate history in the context: first prompt the model as Agent A, then include Agent A's response in a new prompt for Agent B, continue adding each response to the growing debate context, and finally run the synthesis prompt with the complete debate as input. This 'mock debate' approach is less computationally elegant than parallel agent execution but is fully achievable with any single LLM. What tasks are most and least suited for Multi-Agent Debate? Most suited: complex factual claims requiring verification (medical, legal, scientific), high-stakes decisions with significant trade-offs, ethical dilemmas, strategic planning, and any analysis where you want to ensure that the strongest objections to a conclusion have been explicitly considered and answered. Least suited: simple factual lookups, creative writing where convergence is undesirable, time-sensitive tasks where debate latency is prohibitive, and tasks where all relevant information fits in a single well-structured prompt without genuine ambiguity. ## Origin & Sources 1. Du et al. (2023). [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://arxiv.org/abs/2305.14325) . arXiv / ICML 2024. 2. Liang et al. (2023). [Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate](https://arxiv.org/abs/2305.19118) . arXiv. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/neuroprompt [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › NeuroPrompt Advanced Technique Advanced # NeuroPrompt (NP) Apply cognitive neuroscience to prompt design — manage cognitive load, trigger deliberate vs intuitive reasoning modes, and prime attention for complex multi-faceted tasks. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Inspired by Cognitive load theory, dual-process theory Key techniques Chunking, mode-switching, attention priming Underlying research Sweller (1988), Kahneman (2011) Best for Complex analysis, educational content, nuanced reasoning **TL;DR:** NeuroPrompt applies cognitive science principles — cognitive load theory, attention management, and dual-process reasoning — to prompt design, treating the LLM's processing like a human working memory system. Use it for complex, multi-constraint tasks where standard prompts produce degraded or incomplete output due to information overload. Chunk complex tasks into sequential sub-prompts, prime with reasoning mode before the main task, and place critical constraints at the start rather than the end. ## What is NeuroPrompt? NeuroPrompt is a prompting approach that applies principles from cognitive neuroscience and psychology to the design of LLM prompts. The premise is that LLMs, like human cognitive systems, perform better when prompts are structured to match how information processing works — managing the cognitive load of competing requirements, activating the right reasoning mode for the task, and directing attention to critical information before the main work begins. NeuroPrompt draws on three core cognitive science frameworks: - **Cognitive Load Theory (Sweller, 1988):** Working memory is limited. Overloaded prompts — those with too many simultaneous constraints, goals, and context — produce degraded output. NeuroPrompt chunks complex tasks into sequential sub-tasks, each with a single focused objective. - **Dual-Process Theory (Kahneman, 2011):** System 1 reasoning is fast, intuitive, and pattern-driven; System 2 is slow, deliberate, and analytical. NeuroPrompt uses explicit mode-switching language to activate the appropriate reasoning style for each phase of a task. - **Attention Priming:** LLMs weight early context more heavily. By explicitly naming the most critical requirements or constraints at the prompt's opening, you prime the model's "attention" toward what matters most before it begins generating. ## When to Use NeuroPrompt 🧠 ### Complex Cognitive Tasks Multi-dimensional analysis where a single prompt with all requirements produces shallow output on every dimension — cognitive load chunking lets the model fully address each requirement in turn. 📚 ### Educational Content Generation Instructional material benefits enormously from cognitive scaffolding principles: building on prior knowledge, introducing one concept at a time, and checking comprehension incrementally. 🔍 ### Nuanced Multi-Angle Analysis Business decisions, policy analysis, and strategic assessments where you need both creative ideation (System 1) and rigorous analytical validation (System 2) as distinct phases. 🎨 ### Creative Work with Complex Constraints Creative writing or design tasks with many simultaneous constraints benefit from a free ideation phase followed by a constraint-evaluation pass rather than trying to satisfy all constraints during generation. ⚖️ ### Ethical & Risk Analysis Decisions with moral dimensions benefit from a structured multi-perspective approach: generate all considerations first (System 1 scan), then systematically evaluate each (System 2 analysis). 🏋️ ### Skill Development & Tutoring AI tutoring systems that apply cognitive load theory produce more effective instruction — scaffolding complexity progressively and using worked examples before practice problems. ## How to Apply NeuroPrompt Techniques 1. 1 ### Identify the cognitive demands of the task Before writing the prompt, map out what cognitive operations are required: Is this primarily pattern recognition (System 1) or logical analysis (System 2)? How many simultaneous constraints are involved? What information is most critical and might get lost in a long prompt? This analysis guides which NeuroPrompt techniques to apply. 2. 2 ### Open with attention priming State the single most critical constraint or requirement in the first 1–2 sentences of the prompt. Use explicit attentional language: "The most important consideration throughout this task is...", "Pay particular attention to...", "Before you begin, note that...". This anchors the model's processing around what matters most. 3. 3 ### Chunk complex tasks into sequential sub-tasks Break the overall task into 3–5 discrete steps, each with a single objective. Label each step clearly (Step 1, Step 2, or named phases). This prevents cognitive load overload where competing requirements degrade each other. The model completes each sub-task fully before moving to the next. 4. 4 ### Use mode-switching language for reasoning style Explicitly signal which reasoning mode to use at each phase: "First, brainstorm freely without filtering (intuitive scan)..." followed by "Now switch to careful analytical reasoning — for each idea, evaluate it against..." This dual-process structure produces both creative breadth and analytical rigor in the same output. 5. 5 ### Request a structured synthesis After the chunked sub-tasks, ask the model to synthesize findings into a specified output format. The synthesis step mirrors how human working memory consolidates information from multiple cognitive operations into a coherent conclusion — and gives you the clean, actionable output you actually need. ## Prompt Examples Dual-Process Business Analysis with Cognitive Load Chunking ``` ## NeuroPrompt — Cognitive Load Chunking for Complex Analysis ### Attention Prime Before you begin, the most critical requirement is: your analysis must distinguish between short-term and long-term effects. Keep this distinction central throughout. ### Task Decomposition (Chunked for Working Memory) **Step 1 — System 1 Scan (Intuitive Pass)** Read the following business scenario and list your immediate intuitive observations. Don't filter — capture first impressions, patterns you notice, and anything that stands out. (3–5 bullet points) [SCENARIO: A SaaS company is considering dropping its freemium tier to focus exclusively on paid plans. Monthly active users: 50,000. Paid conversion rate: 2%. Average revenue per paid user: $45/month.] **Step 2 — System 2 Analysis (Deliberate Pass)** Now switch to careful analytical reasoning. For each observation from Step 1, either validate it with specific figures from the scenario or discard it if it doesn't hold up under scrutiny. **Step 3 — Structured Synthesis** Present your final analysis in this format: - Short-term impact (0–6 months): [revenue, user, operational effects] - Long-term impact (6–24 months): [strategic, competitive, growth effects] - Recommendation: [one clear decision with rationale] ``` Working Memory Optimization for Educational Content ``` ## NeuroPrompt — Working Memory Optimization for Educational Content You are an expert instructional designer applying cognitive load theory. **Attention Prime:** The learner is a complete beginner. Never assume prior knowledge. Every new concept must be anchored to something familiar before being explained. **Task:** Explain how HTTP requests and responses work. **Cognitive Scaffolding Instructions:** 1. Start with a concrete analogy from everyday life (anchor to familiar experience) 2. Introduce exactly ONE new technical term per paragraph — define it immediately 3. After every two concepts, include a "Check Your Understanding" question that only requires the concepts covered so far (no forward references) 4. Use a worked example that builds incrementally — do not introduce the full example upfront; reveal each part as the relevant concept is explained 5. End with a visual summary (ASCII diagram) showing all concepts and their relationships Maximum paragraph length: 4 sentences. Prioritize clarity over completeness. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dramatically improves output quality on complex, multi-requirement tasks | Prompts become long and complex — higher token cost per interaction | | Produces more cognitively effective educational and explanatory content | Requires significant upfront analysis of the task's cognitive demands | | Dual-process mode switching captures both creative breadth and analytical rigor | The cognitive science analogy is imperfect — LLMs are not human cognitive systems | | Applicable to any LLM without special capabilities or fine-tuning | Overkill for simple, single-requirement tasks where a direct prompt suffices | ## NeuroPrompt Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Focus: this task requires {{ cognitive_demand }}. Read everything before starting. Background (hold this in mind): {{ background }} Work in stages: Stage 1 — Fast pass: give your intuitive first take in 2–3 sentences. Stage 2 — Slow pass: {{ analytical_steps }} Stage 3 — Reconcile: where do the fast and slow passes differ, and which is right? Output: {{ output_format }} ``` ## Frequently Asked Questions What is NeuroPrompt? NeuroPrompt is a prompting approach inspired by cognitive neuroscience that structures prompts to activate specific reasoning modes in LLMs. It draws on principles like cognitive load theory (chunking complex tasks to avoid overwhelming working memory), dual-process theory (distinguishing fast intuitive reasoning from slow deliberate analysis), and attention priming (directing the model's focus to the most relevant information before the main task begins). What is dual-process theory and how does it apply to prompting? Dual-process theory (Kahneman, Thinking Fast and Slow) describes two cognitive systems: System 1 (fast, intuitive, pattern-matching) and System 2 (slow, deliberate, analytical). In prompting, you can trigger System 1-like responses with open-ended, generative instructions ('brainstorm freely') and System 2-like responses with structured, deliberate instructions ('analyze each option systematically before deciding'). NeuroPrompt uses explicit mode-switching language to invoke the appropriate reasoning style for each subtask. What is cognitive load management in prompting? Cognitive load theory (Sweller, 1988) holds that working memory has limited capacity. In prompting, this translates to: avoid giving a model too many simultaneous constraints, requirements, and context — it leads to degraded output quality on all dimensions. NeuroPrompt manages cognitive load by chunking complex tasks into sequential sub-tasks, each with a focused objective, rather than specifying all requirements in a single dense prompt. What is attention priming in the context of LLM prompts? Attention priming is a technique where you explicitly highlight the most critical information or constraints at the beginning of a prompt, before the main task. Research shows LLMs (like humans) weight early context more heavily. By stating 'The most important consideration is X' or 'Pay particular attention to Y' upfront, you effectively prime the model's attention toward what matters most before it processes the task. How is NeuroPrompt different from standard Chain-of-Thought prompting? Chain-of-Thought focuses on making reasoning explicit and sequential. NeuroPrompt goes further by considering which type of reasoning to activate (intuitive vs deliberate), how to structure information to minimize working memory overload, and how to prime attention before reasoning begins. It is a meta-framework for cognitive design of prompts, whereas CoT is a specific technique for eliciting step-by-step reasoning. Is NeuroPrompt grounded in published research? NeuroPrompt as a named framework is a practitioner-developed application of cognitive science principles to prompt engineering, rather than a single peer-reviewed paper. The underlying principles are well-established: cognitive load theory (Sweller, 1988), dual-process theory (Kahneman, 2011), and working memory models (Baddeley, 1992). The prompting applications of these principles are an active area of informal research in the prompt engineering community. What tasks benefit most from NeuroPrompt techniques? NeuroPrompt techniques are most valuable for complex, multi-faceted tasks where output quality is sensitive to how the prompt is framed: nuanced analysis requiring multiple reasoning styles, educational content generation where cognitive scaffolding matters, creative tasks with complex constraints, and any task where standard prompting produces inconsistent or shallow results due to competing requirements overloading a single prompt. ## Origin & Sources 1. Sweller (1988). [Cognitive Load During Problem Solving: Effects on Learning](https://doi.org/10.1207/s15516709cog1202_4) . Cognitive Science 12(2). 2. Kahneman (2011). [Thinking, Fast and Slow](https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow) . Farrar, Straus and Giroux. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/orca [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › ORCA Framework Structured Framework Beginner # ORCA Framework Define your Objective, Role, Context, and Action to give AI models a crystal-clear brief — and get precise, on-target responses every time. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Objective · Role · Context · Action Also known as ORCA Method, O-R-C-A Prompting Best models Any instruction-following model Token cost Low **TL;DR:** ORCA (Objective, Role, Context, Action) is a four-component framework that places the desired outcome first and the specific action last, ensuring every request is anchored in a clear goal before the task is issued. Use it for task-oriented prompts — project deliverables, research tasks, and operational instructions — where scope drift is a risk. State the Objective before the Role to help the model select the most relevant expertise for the goal. ## What is the ORCA Framework? ORCA is a four-component prompt engineering framework that structures every prompt around the four questions that matter most: what do you want to achieve, who should achieve it, what is the situation, and what exactly should happen. The acronym stands for **Objective, Role, Context, and Action**. - **O — Objective:** State what you want to achieve - **R — Role:** Assign the model's persona and expertise - **C — Context:** Provide the background and constraints - **A — Action:** Give the specific instruction The framework is designed for task-oriented prompting — situations where you need the model to do something specific and do it well. By separating the goal (Objective) from the instruction (Action), ORCA forces you to think about the *purpose* of the task before you describe the task itself. This distinction produces prompts that are more coherent and easier for a model to execute than a single instruction block. ORCA sits in the same family as [RACE](https://www.promptedit.app/prompt-framework/race), [RTF](https://www.promptedit.app/prompt-framework/rtf), and [CO-STAR](https://www.promptedit.app/prompt-framework/costar) — structured frameworks that use labeled components to eliminate ambiguity. At four components, ORCA hits a sweet spot: comprehensive enough to cover the full context of a task, simple enough to write quickly without a checklist. ## When to Use the ORCA Framework ✍️ ### Professional Writing Email drafts, reports, summaries, and business documents where the role, goal, and specific task all need to be explicit to get consistent, on-brief output. 📋 ### Task Delegation Prompts When you are briefing an AI assistant on a specific job — research, analysis, drafting — ORCA's structure mirrors how you would brief a human colleague. 📣 ### Marketing & Content Ad copy, social posts, email campaigns — where role (brand voice), context (campaign background), and action (write X words in Y format) all need to be precise. 🔍 ### Structured Analysis When you need the model to analyze data, a document, or a situation from a specific expert perspective with a defined output. 🤖 ### AI Assistant Configuration Use ORCA as a system prompt template to define a focused AI assistant: its mission, persona, operating context, and default behavior. 📐 ### Repeatable Prompt Templates Build reusable prompt templates for recurring tasks — ORCA's four labeled fields are easy to fill in with new values for each use. ## How to Use the ORCA Framework 1. 1 ### Objective — State what you want to achieve Define the goal of the task at the outcome level, not the instruction level. "Write a status update" is an instruction. "Communicate project progress to non-technical stakeholders in a way that surfaces risks and decisions needed" is an Objective. The Objective sets the purpose that the Action must serve. 2. 2 ### Role — Assign the model's persona and expertise Specify who the model is: their title, domain expertise, perspective, and communication style. A well-defined Role activates the appropriate vocabulary, decision-making style, and assumptions. "You are a senior UX researcher who presents findings to product teams" produces very different output than "You are a copywriter." 3. 3 ### Context — Provide the background and constraints Give the model everything it needs to understand the situation: relevant background, the audience, any constraints or requirements, prior work, and domain-specific details. Context is where you transfer knowledge from your head to the model. Thin context produces generic output; rich context produces relevant output. 4. 4 ### Action — Give the specific instruction State exactly what the model should do: produce, write, list, analyze, summarize, compare. Include format requirements (word count, bullet points, sections), structure expectations, and any must-haves or must-avoids. The Action is the direct task; the Objective is why it matters. Both together produce coherent, purposeful output. ## Prompt Examples Professional Writing — Executive Project Status Update ``` Objective: Write a concise executive summary of a software project status update that can be shared with non-technical stakeholders. Role: You are a senior technical program manager who specializes in translating engineering progress into clear business language. You communicate technical complexity without using jargon, and you focus on business impact, risks, and decisions needed. Context: The project is a payment gateway integration for an e-commerce platform. The engineering team has completed the core API connection and sandbox testing. They are now blocked on security compliance review, which is estimated to add two weeks to the timeline. The original deadline was end of Q2; the revised estimate is mid-July. The stakeholders are the CFO and VP of Sales who care about revenue impact and customer commitments. Action: Write a 150-word executive summary covering: (1) what was accomplished this sprint, (2) the current blocker and its cause, (3) the revised timeline, and (4) any decisions or approvals needed from leadership. Use plain language. Lead with the most important information. ``` Marketing — Re-engagement Email Subject Lines ``` Objective: Generate five subject line options for a re-engagement email campaign targeting lapsed newsletter subscribers who have not opened an email in 90 days. Role: You are an email marketing copywriter with expertise in re-engagement campaigns. You know that lapsed subscribers respond best to curiosity, exclusivity, and low-commitment offers. You write subject lines that feel personal and relevant, not promotional. Context: The newsletter covers personal finance and investing for millennials. The brand voice is friendly, direct, and jargon-free — think "your financially savvy friend" rather than "wealth management professional." Previous re-engagement campaigns saw best results with curiosity-driven subject lines and direct "we miss you" messaging. Avoid discount-based subject lines as they attract the wrong segment. Action: Write five subject line options. For each, include a one-sentence note explaining the psychological trigger it uses (curiosity, social proof, FOMO, direct appeal, or personalization). Keep all subject lines under 50 characters. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Separates goal (Objective) from instruction (Action) — produces more purposeful prompts | No explicit output format component — add format requirements inside Action | | Four components cover the full brief without over-engineering | No tone or audience component — extend with notes in Context if needed | | Natural thinking order makes it fast to write and easy to remember | Objective and Action can overlap for simple tasks — ORCA is most valuable for complex briefs | | Works well as a system prompt template for task-focused AI assistants | | ## ORCA Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Objective: {{ objective }} Role: You are {{ role }}. Context: {{ context_and_constraints }} Action: {{ specific_instruction }} ``` ## Frequently Asked Questions What does ORCA stand for in prompt engineering? ORCA stands for Objective, Role, Context, and Action. It is a four-component structured prompting framework that gives an AI model a complete brief in a single, well-organized prompt. Objective defines what you want to achieve, Role sets the persona the model should adopt, Context provides the background and constraints, and Action specifies the exact task to perform. How is ORCA different from CO-STAR? CO-STAR uses six components — Context, Objective, Style, Tone, Audience, and Response — with a strong emphasis on communication style and audience targeting. ORCA is leaner at four components and focuses on task execution: what to achieve, who is doing it, the situation around it, and the specific action to take. ORCA is faster to write and works well for task-oriented prompts; CO-STAR is better when precise tone, style, and audience calibration are the priority. How is ORCA different from RACE? RACE (Role, Action, Context, Expectation) is structurally close to ORCA. The key difference is the fourth component: ORCA ends with Action — a direct instruction for what to do — while RACE ends with Expectation — a description of the desired output format or quality standard. ORCA tends to produce more decisive, task-focused prompts; RACE gives more control over what the output should look like. Is ORCA good for beginners? Yes. ORCA is one of the most beginner-friendly structured frameworks because it follows a natural thinking order: what do I want (Objective), who should do it (Role), what is the situation (Context), and what exactly should happen (Action). Most people can write their first ORCA prompt in minutes without consulting documentation. Can I use ORCA for system prompts? ORCA works well as a system prompt template when you need a task-focused AI assistant. Use Objective to define the assistant's primary mission, Role to establish its persona and expertise, Context to provide domain knowledge and operating constraints, and Action to give standing instructions for how it should respond. For more communication-style control, consider extending ORCA with tone and audience notes, or switching to CO-STAR. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/oscar [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › OSCAR Framework Structured Framework Intermediate # OSCAR Framework Apply project management rigour to your prompts — define Objective, Scope, Constraints, Assumptions, and Results before the AI does a single line of work. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Objective, Scope, Constraints, Assumptions, Results Inspired by Project management methodology Best for Technical tasks, scoped analysis, project work Token cost Moderate to high **TL;DR:** OSCAR (Objective, Scope, Constraints, Assumptions, Results) is a project-management-derived framework that forces explicit scope definition and assumption declaration before any work begins. Use it for product requirements, project briefs, and any deliverable where unstated assumptions or scope creep commonly undermine output quality. The Assumptions component is OSCAR's differentiator — list your working assumptions and ask the model to flag any that seem incorrect. ## What is the OSCAR Framework? OSCAR is a five-part structured prompt framework adapted from project management practice. Its components — Objective, Scope, Constraints, Assumptions, Results — define a task with enough precision that the AI operates within clear boundaries and produces a well-specified deliverable rather than a generic response. - **O — Objective:** Define the goal clearly - **S — Scope:** Specify what is in and what is out - **C — Constraints:** State hard limits the model must respect - **A — Assumptions:** Document what you are taking as given - **R — Results:** Describe the expected deliverable The framework is particularly powerful because it does what most prompts fail to do: it states both what is included and what is excluded (Scope), identifies hard limits (Constraints), and documents what the model should take as given rather than debate (Assumptions). These three elements alone eliminate a large proportion of unhelpful AI responses. Teams that already use project management language will find OSCAR immediately intuitive. For others, the five labels act as a checklist: work through each one and you will have written a better prompt than 90% of freeform attempts. ## When to Use the OSCAR Framework 🏗️ ### Technical Architecture Define system design tasks with explicit technology constraints, excluded components, and a specified deliverable format like a written architecture decision record. 🔍 ### Scoped Research Bound a research task to specific time periods, geographies, or data sources to prevent the AI from ranging too broadly or including irrelevant information. ⚠️ ### Risk Assessment Analyse risk for a defined scope — a specific product, process, or decision — with explicit assumptions about what is already mitigated or out of scope. 📋 ### Project Planning Generate project plans, timelines, or resource breakdowns where budget, technology, and team constraints must be respected throughout. 📈 ### Business Analysis Produce market, competitive, or financial analyses bounded to specific segments, metrics, or periods with clear deliverable formats. 🧪 ### Test Planning Design test strategies within a defined scope, excluding already-covered areas, and constrained by available tooling and time. ## How to Use the OSCAR Framework 1. 1 ### Objective — Define the goal clearly State what you need to achieve in a single, unambiguous sentence. Avoid compound objectives. If you have two goals, write two OSCAR prompts. A strong Objective sounds like a project charter statement: "Evaluate three database options for a high-read, low-write SaaS application expected to scale to 10M users." 2. 2 ### Scope — Specify what is in and what is out Explicitly name both inclusions and exclusions. "Scope includes PostgreSQL, MySQL, and DynamoDB. Scope excludes graph databases and any solution requiring on-premise hosting." Without exclusions, the model will tend to expand scope on its own — often unhelpfully. 3. 3 ### Constraints — State hard limits the model must respect List technical, budgetary, regulatory, or stylistic limits that cannot be compromised. "Must be deployable on AWS. Must comply with GDPR. Monthly cost must not exceed $2,000 at 1M active users. Output must be in plain prose, no code." Constraints are non-negotiables — frame them as must/must not. 4. 4 ### Assumptions — Document what you are taking as given List facts the model should treat as true without challenging them. "Assume the engineering team has strong SQL skills but no NoSQL experience. Assume existing data is fully relational. Assume no migration budget is available." This prevents the model from questioning your premises and keeps it focused on the actual analysis. 5. 5 ### Results — Describe the expected deliverable Specify the output format, structure, and length. "A written comparison in three sections: overview of each option, scored evaluation matrix (criteria: cost, scalability, team fit, operational complexity), and a final recommendation with one paragraph of justification." The more specific your Results definition, the more usable the output. ## Prompt Examples OSCAR — Database Selection Analysis ``` OBJECTIVE: Recommend the best database technology for a new B2B SaaS application with high read volume, moderate write volume, and complex relational data. SCOPE: - In scope: PostgreSQL, MySQL, CockroachDB - Out of scope: NoSQL options, graph databases, on-premise solutions CONSTRAINTS: - Must run on managed cloud service (AWS RDS or equivalent) - Monthly cost must not exceed $1,500 at 500k active users - Must support ACID transactions - Output must be written prose, no code samples ASSUMPTIONS: - Engineering team is proficient in SQL but has no NoSQL experience - All existing data is relational with clear foreign key relationships - The application will be deployed on AWS RESULTS: A three-section written analysis: (1) brief overview of each option, (2) evaluation against the four criteria: cost, scalability, team fit, ACID compliance — using a clear scoring narrative, (3) a final recommendation with two paragraphs of justification. ``` OSCAR — Competitive Market Analysis ``` OBJECTIVE: Analyse the competitive landscape for AI-powered legal contract review tools in the UK small business market. SCOPE: - In scope: UK market, tools targeting SMBs with under 50 employees, products launched or updated in the last 3 years - Out of scope: enterprise legal platforms, US-only products, human-staffed legal services CONSTRAINTS: - Analysis must be based on publicly available information only - No speculation about private company financials - Output must be under 600 words - Identify no more than 5 competitors ASSUMPTIONS: - The reader is a founder preparing a pitch deck, not a legal expert - Pricing models are publicly listed on competitor websites - AI-generated contract review is legally permissible in the UK (do not debate this point) RESULTS: A structured competitive brief with: (1) a one-paragraph market overview, (2) a competitor table listing name, key feature, pricing model, and apparent target customer, (3) a gap analysis identifying two underserved segments or unmet needs. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Explicit scope boundaries prevent AI from drifting off-topic | More upfront investment than simpler one-paragraph prompts | | Assumptions section keeps the model focused rather than questioning premises | Five sections can feel bureaucratic for creative or conversational tasks | | Natural language for anyone familiar with project management | Requires clear thinking about boundaries before you start — not ideal for exploratory work | | Highly reusable — swap Objective and Input for similar tasks | | ## OSCAR Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Objective: {{ objective }} Scope: In scope: {{ in_scope }}. Out of scope: {{ out_of_scope }}. Constraints: {{ hard_limits }} Assumptions: {{ assumptions }} Results: {{ expected_deliverable }} ``` ## Prompt Flows That Use OSCAR Framework OSCAR Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Sprint Planner](https://www.promptedit.app/prompt-flow/the-sprint-planner) · Project Planning ## Frequently Asked Questions What does OSCAR stand for in prompting? OSCAR stands for Objective, Scope, Constraints, Assumptions, Results. Borrowed from project management, it gives the AI a complete picture of what success looks like, what is in and out of bounds, what cannot be changed, what can be taken as given, and what the deliverable should be. Together these five elements prevent the vague, over-broad outputs that plague unstructured prompts. Where did the OSCAR framework come from? OSCAR originated in project management and systems engineering as a requirements-definition tool. Prompt engineers adapted it for AI because the same information that scopes a project also scopes a model's behaviour — without a defined objective, scope, and constraints, both projects and prompts drift. How is OSCAR different from SMART goals? SMART (Specific, Measurable, Achievable, Relevant, Time-bound) focuses on goal quality. OSCAR goes further by adding Scope (what is excluded), Constraints (hard limits), Assumptions (what we take as given), and Results (deliverable format). OSCAR is more complete for complex analytical tasks that need explicit boundary-setting. When should I choose OSCAR over simpler frameworks? Choose OSCAR when the task has real complexity: technical analysis, scoped research, architecture decisions, risk assessments, or any deliverable where the AI could legitimately interpret the brief in several different ways. If your task is simple and the output format is obvious, a lighter framework will do. What goes in the Assumptions section? Assumptions are things you are treating as true even though they could be challenged — and you want the model to work within them rather than question them. Examples: 'Assume the technology stack is fixed and cannot change', 'Assume the budget is $50k', 'Assume the audience has intermediate Python knowledge'. Listing assumptions also makes your reasoning auditable. Can OSCAR be used for creative tasks? OSCAR can be applied creatively, but it is most natural for analytical, technical, or project-oriented work. For purely creative tasks, frameworks like SCAMPER or role prompting tend to feel less restrictive. That said, constrained creative briefs — writing a short story within strict genre, length, and content requirements — benefit from OSCAR's boundary-setting. How do I define the Results section effectively? The Results section should describe the deliverable format, not just the topic. Instead of 'a summary', specify 'a 3-section written report with an executive summary, detailed findings, and a table of recommended actions'. The clearer the output specification, the less formatting cleanup you will need after generation. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/pas [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › PAS Framework Copywriting Beginner # PAS Framework The simplest persuasion engine in copywriting — identify the Problem, intensify it with Agitation, then deliver the Solution that earns the sale. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Problem, Agitate, Solution Best for Sales copy, landing pages, cold audiences Token cost Low **TL;DR:** PAS (Problem, Agitate, Solution) is a three-beat copywriting formula that names a pain point, amplifies the emotional cost of ignoring it, then presents your offer as the relief. Use it for sales emails, landing page hero sections, and ad copy where moving the reader from problem awareness to purchase intent is the explicit goal. The Agitate stage does the heavy lifting — be specific about consequences, not just vague discomfort. ## What is the PAS Framework? PAS — Problem, Agitate, Solution — is one of the most widely used and consistently effective formulas in copywriting. In three beats, it moves a reader from awareness of a problem to emotional motivation to act, to relief through your offer. Marketing legend Dan Kennedy called PAS "the most reliable sales formula ever invented." - **P — Problem:** Identify and name the pain point - **A — Agitate:** Intensify the emotional weight - **S — Solution:** Present the relief clearly and confidently For AI prompting, PAS is particularly powerful because its three-step structure is unambiguous. When you tell a model to write using PAS, it knows exactly what each section must do: identify pain, deepen the emotional stakes, deliver the answer. The result is persuasive copy with a natural arc rather than a flat feature list dressed up as marketing. Unlike AIDA, which often leads with a hook designed to create curiosity, PAS leads with the reader's existing pain. This makes PAS especially effective for audiences who already know they have a problem — cold email, search ads, retargeting campaigns, and sales pages where the reader has arrived because they are already looking for a solution. ## When to Use the PAS Framework 🛒 ### Sales Copy Write product and service sales copy that earns trust by naming the customer's pain before pitching the offer. 🌐 ### Landing Pages Structure the hero section of a landing page so visitors immediately recognise their problem and feel compelled to read on. 📧 ### Cold Email Open B2B outreach by demonstrating you understand the prospect's pain before making any mention of your solution. 📢 ### Search & Retargeting Ads Match the pain-point language your audience is already using in search queries and bring them to a solution that feels tailor-made. 📱 ### Social Media Posts Open with a painfully relatable scenario, agitate briefly, and pivot to a solution — the structure works in 150 words or 500 words alike. 📣 ### Video Scripts Script YouTube pre-rolls, explainer videos, and testimonial compilations that follow the problem-agitate-solution arc viewers have been conditioned to expect. ## How to Use the PAS Framework 1. 1 ### Problem — Identify and name the pain point State the reader's core problem as specifically as possible. Generic problems create generic copy. "You're losing revenue" is weak. "You're spending 3 hours every Friday manually reconciling invoices that your accountant still gets wrong" is strong. In your prompt, give the AI the specific problem your audience faces — do not leave this to the model to invent. 2. 2 ### Agitate — Intensify the emotional weight Make the problem feel urgent and consequential. Describe what happens if it goes unsolved: the compounding costs, the hidden toll, the risks, the opportunity being lost. Agitation is not cruelty — it is empathy made specific. Tell the model to amplify the stakes, mention secondary consequences, and help the reader feel the full cost of inaction. 3. 3 ### Solution — Present the relief clearly and confidently Now reveal your offer as the natural, obvious answer to everything the previous two sections built up. The Solution should feel like exhaling. Describe the outcome, not just the mechanism. Tell the model to transition smoothly from the agitated state to relief, and close with a specific call to action. ## Prompt Examples PAS — SaaS Landing Page Hero Section ``` Write a landing page hero section using the PAS framework for a project management tool aimed at agency owners. PROBLEM: Name the specific pain of agencies losing track of project scope, deadlines slipping, and clients being surprised by delays they should have known about weeks ago. AGITATE: Expand on the consequences — team morale, client churn, reputation damage, and the founder personally fire-fighting instead of growing the business. Make it feel real and personal. SOLUTION: Introduce the tool as the system that gives agency owners a real-time view of every project, automates client status updates, and surfaces problems before they become crises. End with a CTA for a free 14-day trial. Total length: 180–220 words. Direct, empathetic, no fluff. ``` PAS — B2B Cold Email Opening ``` Write the opening three paragraphs of a cold email using PAS. Target: Operations managers at e-commerce companies shipping over 1,000 orders per month. PROBLEM: They are spending hours each week manually processing returns and refunds because their current system has no automation. AGITATE: The downstream effects — customer complaints about slow refund processing, customer service team overload, increasing churn from a poor post-purchase experience, and the operations manager being pulled into tactical work instead of process improvement. SOLUTION: Tease (do not fully explain) a returns automation platform that handles the full refund workflow in under 60 seconds per return, with zero manual input. Close the opening with a soft CTA asking if they have 15 minutes this week to see it in action. Keep the total under 130 words. Confident, peer-to-peer tone — not salesy. No subject line needed. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Extremely simple — three steps anyone can understand and apply immediately | Agitation can feel manipulative if the problem is exaggerated or fabricated | | Highly effective with cold audiences who already recognise the problem | Less effective when the audience does not yet know they have the problem | | Works at any length, from a single sentence to a full sales page | The three-beat structure can become predictable to sophisticated readers | | Builds empathy before making any pitch, which increases trust | | ## PAS Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Write {{ content_type }} for {{ product }} aimed at {{ audience }}. Problem: {{ pain_point }} Agitate: Describe the cost of ignoring it: {{ consequences }} Solution: Present {{ product }} as the fix, with {{ proof_point }}. ``` ## Prompt Flows That Use PAS Framework PAS Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Market Maker](https://www.promptedit.app/prompt-flow/the-market-maker) · Marketing Copy ## Frequently Asked Questions What is the PAS framework? PAS stands for Problem, Agitate, Solution. It is a three-step copywriting formula that identifies a reader's pain point (Problem), intensifies their awareness of how bad that pain really is (Agitate), and then presents your offer as the relief they have been looking for (Solution). PAS is considered one of the most effective short-form persuasion structures ever devised. What makes PAS so powerful for short-form copy? PAS works in as little as three sentences because every step earns the next. The Problem creates relevance — the reader recognises themselves. The Agitate creates urgency — the reader feels motivated to fix it. The Solution provides relief — the reader has a clear path forward. This three-beat psychological rhythm is highly efficient and deeply persuasive. When should I use PAS instead of AIDA? PAS is better when your strongest asset is your understanding of the reader's pain — when you can describe their problem better than they can. AIDA is better when you lead with a powerful hook or brand story and need to build desire before revealing the solution. PAS is often more effective for cold audiences because it demonstrates empathy before making any pitch. What does 'Agitate' mean in practice? Agitation is the step that separates PAS from a simple problem-solution pitch. Agitation makes the reader feel the full weight of the problem — its consequences, hidden costs, emotional toll, or the compounding effect of ignoring it. It is not about being cruel; it is about creating the emotional energy needed to motivate action. Without agitation, problems feel abstract and solutions feel optional. Can PAS be used for B2B content? Yes. B2B buyers are still humans and still respond to pain-point-first messaging. In B2B, the Problem is often operational (lost time, compliance risk, team inefficiency) and the Agitation focuses on business consequences (cost, competitive disadvantage, executive visibility). The emotional register is more professional but the PAS structure works identically. How long should each PAS section be? For short-form copy (ads, email intros, social posts): one to three sentences each. For long-form sales pages: Problem can be a full section with examples, Agitate can include multiple consequences and scenarios, and Solution expands to feature details, pricing, testimonials, and FAQs. The ratio roughly stays 20% Problem, 30% Agitate, 50% Solution for long-form. Is PAS ethical? Does agitating pain manipulate readers? PAS is ethical when the Problem you describe is real and the Solution you offer genuinely addresses it. Naming a real pain and offering a real solution is the basis of all useful commerce. PAS becomes manipulative only if the Problem is fabricated, the Agitation is exaggerated beyond truth, or the Solution cannot actually deliver what it promises. Use it honestly and it is one of the most reader-respecting frameworks available. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/pecra [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › PECRA Framework Structured Framework Intermediate # PECRA Framework Craft audience-first content by locking in Purpose, Expectation, Context, Request, and Audience before asking the AI to write a single word. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Purpose, Expectation, Context, Request, Audience Best for Marketing, communications, stakeholder content Token cost Moderate **TL;DR:** PECRA (Purpose, Expectation, Context, Request, Audience) is an audience-first framework that forces you to articulate why the content must exist and who must be moved by it before issuing the actual request. Use it for persuasive content, stakeholder communications, and marketing copy where misalignment between content purpose and audience needs is the primary failure mode. State the Audience last, immediately before the Request, so it shapes the model's final word choices directly. ## What is the PECRA Framework? PECRA is a five-element prompt structure that places the reader at the centre of every content decision. Its five components — Purpose, Expectation, Context, Request, Audience — work together to ensure that an AI model understands not just what to write, but why the content exists and who must be moved by it. - **P — Purpose:** Why does this content need to exist? - **E — Expectation:** What does a successful output look like? - **C — Context:** What background does the model need? - **R — Request:** What specifically do you want? - **A — Audience:** Who will read this? Most prompt failures occur because one of these five elements is missing. A prompt with a clear Request but no Audience produces technically correct content that misses the reader's vocabulary, concerns, or emotional state. A prompt with clear Purpose but no Expectation generates the right idea in the wrong format. PECRA closes all five gaps at once. The framework is particularly powerful for marketing, PR, and communications teams who need to produce multiple versions of the same message for different stakeholder groups — each version can share the same Purpose and Context but vary the Audience and Expectation sections. ## When to Use the PECRA Framework 📣 ### Marketing Copy Produce ads, landing pages, and product descriptions that speak directly to a defined customer segment's motivations and language. 📧 ### Email Campaigns Write nurture sequences and promotional emails where tone, urgency, and vocabulary shift based on where the reader is in the funnel. 📰 ### Press Releases Frame company news for journalists, investors, or end customers — each audience needs a different angle and level of detail. 🤝 ### Stakeholder Updates Translate project status into language appropriate for board members, clients, or technical teams without losing the core message. 🌐 ### Website Content Write homepage copy, about pages, and feature descriptions calibrated to the specific buyer persona visiting each section. 📱 ### Social Media Content Adapt a single message to platform-specific audiences — LinkedIn professionals, Instagram consumers, or X power-users — with consistent brand voice. ## How to Use the PECRA Framework 1. 1 ### Purpose — Why does this content need to exist? State the business or communication objective in one or two sentences. "To persuade hesitant prospects to book a demo" is a Purpose. "Write something about our product" is not. A clear Purpose guides every subsequent decision the model makes. 2. 2 ### Expectation — What does a successful output look like? Define format, length, tone, structure, and any non-negotiables. "A 250-word LinkedIn post, professional but warm, ending with a question to drive comments, no bullet points" is a strong Expectation. Include what to avoid as well as what to include. 3. 3 ### Context — What background does the model need? Provide the situational information that shapes the content: industry, company stage, recent events, competitive landscape, or the specific product being discussed. Context prevents generic outputs and grounds the model in your actual situation. 4. 4 ### Request — What specifically do you want? State the explicit deliverable clearly and directly. After all the framing, this is the imperative: "Write a 3-email onboarding sequence" or "Rewrite the hero headline for our pricing page." Keep it unambiguous and action-oriented. 5. 5 ### Audience — Who will read this? Describe your reader with enough specificity to change the output: their role, industry, level of expertise, pain points, and what they care about. "Small business owners in professional services, time-poor, sceptical of tech, primary concern is cost" will produce a very different result than "enterprise IT buyers." ## Prompt Examples PECRA — Product Launch Email ``` PURPOSE: To generate excitement and drive early sign-ups for our new AI-powered invoice tool launching next month. EXPECTATION: A single promotional email, 200–250 words, conversational and enthusiastic tone, subject line included, ending with a prominent CTA button label. No jargon. One short paragraph per idea. CONTEXT: We are a 3-year-old B2B SaaS company targeting freelancers and solo consultants. Our existing users love us for simplicity. The new invoice tool auto-categorises expenses and drafts invoices from calendar entries. Early adopters get 3 months free. REQUEST: Write the launch announcement email. AUDIENCE: Freelance designers, writers, and consultants aged 28–45 who already use our platform. They are comfortable with tech but not developers. Their biggest pain point is the time they lose on admin instead of billable work. ``` PECRA — Investor Update ``` PURPOSE: To reassure Series A investors after a slower-than-projected Q2 and maintain their confidence ahead of the next funding round. EXPECTATION: A 400-word written update (not a slide deck), formal but direct tone, structured as: headline metric, honest explanation, corrective actions taken, forward-looking confidence statement. No spin — investors will see through it. CONTEXT: We missed Q2 ARR target by 18% due to a delayed enterprise deal that has since closed in July. Churn remained at 2.1% MoM (industry average is 3.5%). We hired a new VP of Sales in June. Pipeline is at an all-time high entering Q3. REQUEST: Draft the Q2 investor update letter. AUDIENCE: 6 Series A investors — all experienced SaaS operators who have seen companies navigate rough quarters before. They value honesty and specific data over optimism. They will ask hard questions if the narrative feels incomplete. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Bakes audience intelligence directly into the prompt structure | Five sections require more upfront thinking than simpler frameworks | | Eliminates the most common causes of off-target AI content | Overkill for simple one-off tasks with a well-understood audience | | Easy to adapt: swap Audience to get multiple versions of one message | Audience descriptions can become stereotyped if not researched carefully | | Works across every channel — email, social, PR, internal comms | | ## PECRA Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Purpose: {{ why_this_content_exists }} Expectation: {{ what_success_looks_like }} Context: {{ background }} Request: {{ specific_request }} Audience: {{ audience }} ``` ## Prompt Flows That Use PECRA Framework PECRA Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Market Maker](https://www.promptedit.app/prompt-flow/the-market-maker) · Marketing Copy ## Frequently Asked Questions What does PECRA stand for? PECRA stands for Purpose, Expectation, Context, Request, Audience. It is a five-part prompt framework designed to ensure AI-generated content is precisely targeted at the right readers, for the right reason, in the right format. Each element eliminates a different type of ambiguity that causes AI outputs to miss the mark. How does PECRA differ from PREP or other argumentation frameworks? PECRA focuses on audience-aware content creation — it bakes the target reader into the prompt structure from the start. Frameworks like PREP focus on argumentation structure. PECRA is better when the same message needs to land differently with different audiences: investors vs. customers vs. engineers, for instance. When is PECRA most useful? PECRA is most useful for marketing copy, communications strategy, stakeholder updates, PR materials, and any content where knowing who will read it changes how it should be written. If your prompt does not currently mention the audience, PECRA will immediately improve your output quality. Can I use PECRA for internal communications? Absolutely. PECRA works just as well for internal memos, board updates, team announcements, and change-management communications. Specifying the Audience as 'frontline retail staff with no finance background' versus 'the CFO and finance leadership team' produces dramatically different and more appropriate outputs. What should I put in the Expectation section? The Expectation section describes what a successful output looks like — format, length, tone, structure, and any specific deliverables. Think of it as the brief a creative director gives a copywriter: 'I want a 200-word email, conversational tone, ending with a clear call-to-action, no jargon.' The more precise you are, the less revision you will need. Is PECRA suitable for non-native English speakers using AI writing tools? Yes. PECRA's structured labels make it very easy to use even when you are not fluent in prompt-engineering terminology. Each label is a clear question: 'Why does this content exist? What should it look like? What is happening around it? What do I specifically want? Who will read it?' Anyone can fill in those five answers. How do I handle multiple audiences in a single PECRA prompt? Either create separate PECRA prompts for each audience segment, or add a nuanced Audience section that describes the primary and secondary readers and how their needs differ. For very different audiences, separate prompts usually produce better results than a single prompt trying to serve everyone. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/pivo [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › PIVO Framework Structured Framework Intermediate # PIVO Framework Combine strategic problem framing, data-driven insights, brand voice, and a clear outcome to produce content that is both intellectually grounded and unmistakably on-brand. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Problem · Insights · Voice · Outcome Signature feature Insights layer grounds content in real data Best for Content strategy, thought leadership, brand comms Complexity Medium — 4 strategic components **TL;DR:** PIVO (Problem, Insights, Voice, Outcome) bridges strategic insight and brand expression by combining a defined challenge and supporting data with explicit voice guidance and a concrete deliverable. Use it for brand content, data-driven marketing, and thought leadership where generic writing is unacceptable and voice consistency matters. The Insights component is PIVO's differentiator — supply real data, research findings, or customer quotes to ground the output in evidence. ## What is the PIVO Framework? PIVO is a prompt engineering framework built around four components: **Problem**, **Insights**, **Voice**, and **Outcome**. It bridges strategic thinking and brand expression — combining the "why" (Problem and Insights) with the "how it should sound" (Voice) and "what you need" (Outcome). - **P — Problem:** Define the Challenge - **I — Insights:** Bring Your Knowledge - **V — Voice:** Specify the Brand or Communication Style - **O — Outcome:** State the Desired Result The Insights component is what makes PIVO distinctive. Rather than asking the AI to generate content from scratch, PIVO requires you to bring relevant data, research, or observations into the prompt. The AI's job is then to synthesise those insights into content that carries your brand voice and achieves a specific outcome — not to invent the intellectual substance from nothing. This makes PIVO especially powerful for thought leadership, content strategy, and brand communications where the credibility of the content depends on real knowledge being woven into it. ## When to Use the PIVO Framework ✍️ ### Thought Leadership Turn proprietary data, client observations, and hard-won expertise into published articles, LinkedIn posts, and keynote narratives that establish genuine authority. 📣 ### Brand Communications Ensure every communication — from press releases to customer emails — reflects both the strategic context and the specific voice that defines your brand. 📊 ### Content Strategy Build content strategies grounded in audience insights and competitive data, then specify the voice and outcomes so all content is strategically aligned from the start. 🚀 ### Product Launches Frame the launch around the real problem it solves, leverage customer research insights, and ensure messaging carries a consistent voice across all launch assets. 🤝 ### Client-Facing Reports Produce consulting reports, agency proposals, and client strategy documents that reflect both deep knowledge and a professional, relationship-aware voice. 🎙️ ### Executive Ghostwriting Capture an executive's point of view, blend in supporting data, and produce polished speeches, articles, and social content that authentically represent their voice. ## How to Use the PIVO Framework 1. P ### Problem — Define the Challenge State the specific problem you are trying to solve. This is not the content brief — it is the underlying business or communication challenge that motivated the task. Being precise about the Problem helps the AI understand the stakes and ensures the output is genuinely useful rather than generically relevant. 2. I ### Insights — Bring Your Knowledge Share the relevant data points, research findings, customer feedback, or competitive observations that should inform the output. These are the intellectual inputs — what you know that the AI should weave into its response. The richer your Insights, the more original and credible the output will be. 3. V ### Voice — Specify the Brand or Communication Style Describe how the output should sound. Include tone adjectives, reference brands or publications as style benchmarks, note what to avoid, and describe the relationship the voice suggests (peer, expert advisor, mentor). Voice guidance prevents generic, corporate-sounding output and ensures brand consistency. 4. O ### Outcome — State the Desired Result Describe the specific deliverable and what it should achieve. Include format, length, sections, and any structural requirements. The Outcome is both the deliverable spec and the success criterion — what does this content need to accomplish for the audience? ## Prompt Examples SaaS Trial-to-Paid Conversion Strategy ``` Problem: Our SaaS product has strong retention among power users but struggles to convert free trial users to paid plans. Trial-to-paid conversion sits at 4.2% against an industry benchmark of 8–12%. Exit surveys show users understand the product's features but don't see the immediate value for their specific workflow. Insights: - Users who complete the guided onboarding flow convert at 11.8% (3x the average) - The most common trial action is importing data; users who import and run one report convert at 9.4% - "Aha moment" data shows users who reach their first insight within 48 hours are 6x more likely to convert - Competitor analysis: top competitors lead with use-case-specific landing pages and personalised onboarding tracks Voice: Authoritative but approachable — we are a trusted data partner, not a pushy vendor. Copy should feel like advice from a knowledgeable colleague, not a sales deck. We avoid urgency tactics, discounts, and fear-based messaging. Outcome: A complete content strategy for improving trial-to-paid conversion, including: a revised onboarding email sequence (5 emails), two LinkedIn thought leadership article outlines positioned around the insights above, and messaging guidelines for the in-app upgrade prompt at the moment users run their first report. ``` Consultancy Thought Leadership Platform ``` Problem: Our B2B consultancy is invisible in a crowded market. We have 12 years of client results, but no consistent point of view that differentiates us from generalist strategy firms. Senior partners are reluctant to publish because they fear sharing proprietary thinking. Insights: - Our best client outcomes share a pattern: we challenge the client's initial problem framing before proposing solutions (we call this "reframe before resolve") - 78% of our project work starts with a client coming to us with the wrong problem - Industry publications consistently undercover the cost of misdiagnosed business problems — there is a clear editorial gap we can own - Three clients have publicly credited our reframing approach in case studies but we have never built content around it Voice: Intellectually rigorous, direct, and occasionally contrarian. We write like McKinsey but speak like a trusted advisor at dinner — formal enough to be credible, human enough to be memorable. No jargon without definition. No filler. Every sentence earns its place. Outcome: A thought leadership platform brief including: 1. A core point-of-view statement (2–3 sentences) we can use across all content 2. Three flagship article concepts with working titles and argument outlines 3. A LinkedIn content calendar for one quarter (12 posts, weekly cadence) 4. Guidance on how partners can contribute without feeling exposed ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Insights component produces content grounded in real knowledge, not AI invention | Requires genuine insights to bring — outputs suffer if the Insights section is thin | | Voice specification ensures consistent brand tone across all outputs | More demanding upfront preparation than simpler frameworks | | Strong framework for thought leadership, strategy, and high-value communications | Less suited for tactical, transactional, or data-processing tasks | | Problem framing focuses the AI on what actually matters, not surface-level execution | | ## PIVO Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Problem: {{ problem }} Insights: {{ what_you_know }} Voice: {{ brand_voice }} Outcome: {{ desired_result }} ``` ## Prompt Flows That Use PIVO Framework PIVO Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Brand Voice Engineer](https://www.promptedit.app/prompt-flow/the-brand-voice-engineer) · Brand Strategy ## Frequently Asked Questions What does PIVO stand for? PIVO stands for Problem, Insights, Voice, and Outcome. Problem defines the challenge to solve, Insights bring in relevant data and knowledge that should inform the response, Voice specifies the brand or communication style, and Outcome describes the desired result or deliverable. When should I use the PIVO framework? Use PIVO for content strategy, brand communications, thought leadership, and any task where both the reasoning behind the content and the voice it should be written in are important. PIVO is particularly strong when you have data or research that should shape the output but also have clear brand or tone requirements. What makes PIVO different from other content frameworks? PIVO uniquely combines strategic reasoning (Problem and Insights) with brand expression (Voice and Outcome). Most frameworks focus on either instructions or persona, but PIVO treats the insight layer as essential input — the AI must understand not just what to produce, but what knowledge should be woven into it and what tone it must carry. How specific should the Voice component be? The more specific, the better. Instead of 'professional tone', describe the voice in terms of how it feels: 'like a trusted advisor at dinner — formal enough to be credible, human enough to be memorable'. You can also reference a publication, author, or brand as a benchmark. The goal is to give the AI enough texture to make consistent style decisions. What belongs in the Insights section? Insights should contain data points, research findings, customer feedback, competitive observations, or any knowledge that should actively shape the content. These are not background context — they are the intellectual raw material the AI should weave into the output. Think of it as briefing the AI on what you know before asking it to write. Can PIVO be used for social media content? Yes — PIVO works well for social media, especially for brand accounts with a distinct voice and a point of view to communicate. The Insights component ensures posts are grounded in real data rather than generic platitudes, and the Voice component ensures consistency across different writers or AI-generated drafts. How does PIVO compare to the DEEP framework? Both PIVO and DEEP address brand voice and direction. PIVO leads with the strategic Problem and Insights, making it stronger for thought leadership and content strategy where the 'why' matters. DEEP focuses more on the writing execution — Direction, Existing Info, Expertise, Preferred Tone — making it a better fit for individual content pieces where the strategy is already set. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/prep [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › PREP Framework Communication Framework Beginner # PREP Framework Four tight beats — Point, Reason, Example, Point — that turn any argument into a clear, memorable, and persuasive communication: from three-sentence emails to full presentations and speeches. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Point · Reason · Example · Point (restate) Best for Structured arguments, presentations, persuasive writing, speeches Complexity Low — one of the fastest frameworks to learn and apply Token cost Low **TL;DR:** PREP (Point, Reason, Example, Point) is a structured argumentation framework that bookends evidence with the main claim, making persuasive messages more memorable and logically tight. Use it for executive presentations, debate preparation, persuasive essays, and any content where a clear, defensible position must be communicated quickly. The double Point repetition is intentional — state the claim first to prime the reader, then restate it after the evidence to make it stick. ## What is the PREP Framework? PREP is a four-step communication framework that structures any argument or persuasive message into four components: **Point** (your main claim), **Reason** (why it is true), **Example** (concrete proof or illustration), and **Point** (restatement to reinforce the message). The framework's defining feature is the deliberate repetition of the main point — once to introduce it and once to land it after the evidence has done its work. - **P — Point:** State your main claim directly - **R — Reason:** Explain the primary logical justification - **E — Example:** Support the Reason with concrete proof - **P — Point:** Restate and reinforce your claim Originally developed as a speech and debate training tool, PREP has become one of the most widely taught frameworks in business communication, public speaking coaching, and professional writing. Its power lies in its constraint: by limiting the structure to one reason and one example, PREP forces clarity. It prevents the diffuse, multi-point arguments that bury the main message in a list of supporting claims that no audience can remember. In AI prompting, PREP is particularly effective because it forces the model to commit to a clear position from the first sentence, rather than the hedged, exploratory language that unguided AI often produces. By providing your Point, Reason, and Example explicitly, you give the AI the raw material to construct a tight, coherent argument without filling in the gaps with generic claims. ## When to Use the PREP Framework 🎤 ### Speeches and Presentations Structure individual argument segments within a presentation so each point is made crisply, supported, and reinforced — giving audiences a clear takeaway from every section. ✍️ ### Persuasive Writing Write op-eds, opinion articles, position papers, and thought leadership pieces that take a clear stand and back it with evidence, rather than exploring multiple sides without conclusion. 📧 ### Business Arguments in Email Make a clear, supported case in a professional email without the rambling context-setting that often buries the actual ask. PREP in three to five sentences is more persuasive than three paragraphs of background. 💼 ### Meeting Preparation Prepare tight, structured verbal arguments for meetings and presentations where you need to advocate for a decision, resource, or change of direction in limited time. 🛒 ### Sales and Marketing Copy Write value propositions, landing page copy, and sales emails where a single clear benefit, supported by proof, and restated as a conclusion produces higher conversion than multi-point feature lists. 🎓 ### Academic and Structured Essays Use PREP cycles as the building blocks of structured academic or professional essays — each paragraph becomes one PREP cycle addressing one dimension of the overall argument. ## How to Use the PREP Framework 1. 1 ### Point — State your main claim directly Open with your central argument in a single, declarative sentence. No preamble, no context-setting, no "it depends." A strong Point is specific and takes a clear position that any reader can agree or disagree with. "Async communication makes distributed teams more productive than synchronous-first cultures" is a Point. "There are arguments on both sides of remote work" is not. Give this to the AI and tell it to lead with it — not bury it in paragraph two. 2. 2 ### Reason — Explain the primary logical justification State why your Point is true. PREP works best with one strong, central reason per cycle rather than a list — multiple reasons dilute focus and reduce memorability. "Because asynchronous communication forces people to articulate their thinking clearly in writing, eliminating the meeting overhead that consumes 30% of the average knowledge worker's week" is a Reason. Tell the AI your reason explicitly so it does not substitute a weaker one of its own choosing. 3. 3 ### Example — Support the Reason with concrete proof Provide the specific evidence that substantiates your Reason. Supply real data, studies, named examples, or case studies if you have them — the AI can frame them correctly within the argument. If you do not have evidence, instruct the AI to draw on credible research in the domain and to be specific: named studies, concrete percentages, real company names rather than vague generalisations like "many organizations have found." 4. 4 ### Point — Restate and reinforce your claim Return to your opening Point with a restatement that carries more weight now that the Reason and Example have been presented. The closing Point should feel like a conclusion, not a repetition — slightly reworded, delivered with confidence, and optionally linked to a call to action or implication for the reader. Instruct the AI to make this feel decisive rather than circular. ## Prompt Examples Speech Argument — Four-Day Working Week Pilot ``` You are a professional speechwriter helping a team leader prepare remarks for a company all-hands meeting. Use the PREP framework to structure a 2-minute spoken argument for the following position: Point: Our team should adopt a four-day working week pilot for the next quarter. Reason: Most knowledge workers reach their cognitive limit after six focused hours — the additional time is not productive, it is fatiguing, and that fatigue compounds across the week. Example: The 2019 Microsoft Japan trial of a four-day week found a 40% increase in productivity. The Iceland government trials (2015–2019) found maintained or improved productivity in 85% of participating workplaces. Our own internal data shows that 70% of our team's highest-rated work is completed before 3pm. Point (restate): A four-day week pilot is not a benefit — it is a focused productivity experiment with a clear hypothesis and measurable outcomes. Format: Write this as natural spoken language — not bullet points. Conversational and confident, approximately 250 words. End with a specific call to action asking leadership to approve a 90-day pilot. ``` Homepage Copy — SaaS Value Proposition ``` You are a B2B copywriter helping a SaaS company write a persuasive homepage value proposition. Use the PREP framework to write a tight, punchy value proposition block for the following product: Point: Most project management tools create more overhead than they eliminate. [Product name: Fieldwork] is different — it is built for how field service teams actually work. Reason: Generic project tools require field teams to adapt their work to the software's logic. Fieldwork adapts to the team — with offline-first access, job-site photo capture, and real-time progress that syncs the moment connectivity returns. Example: Crestline HVAC reduced job completion paperwork time by 60% in the first month. Summit Electrical eliminated 90% of end-of-day phone check-ins. Both teams were fully operational on Fieldwork within 48 hours of onboarding. Point (restate): Fieldwork gives field service teams back the time they spend managing software, so they can spend it managing jobs. Format: Write this as three distinct copy blocks for a homepage section: a headline (under 10 words), a subheadline (one sentence), and a body paragraph (60–80 words). No bullet points. Conversion-focused, benefit-led language. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Forces argument-first writing that is immediately legible and credible | Single-reason structure can feel thin for highly complex, multi-faceted arguments | | Beginner-friendly — four intuitive steps anyone can learn in minutes | Requires the user to commit to a clear position — not suited for exploratory writing | | Repeated Point makes arguments memorable and resistant to rebuttal | Evidence quality entirely depends on what the user supplies or what the AI knows | | Scales from a three-sentence email to a full position paper or speech | Can become formulaic if every paragraph in a long piece follows the same pattern | | Prevents the hedged, exploratory output that unguided AI often produces | | ## PREP Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Write a {{ format }} for {{ audience }} using the PREP structure. Point: {{ main_claim }} Reason: {{ primary_reason }} Example: {{ supporting_example }} Point: Restate the claim as a clear call to {{ desired_action }}. ``` ## Frequently Asked Questions What does PREP stand for in prompt engineering? PREP stands for Point, Reason, Example, Point. It is a four-step communication framework where you state your main point, explain the primary reason it is true, support it with a concrete example, and then restate the point to reinforce it. The repeated Point bookends the argument and ensures the main message is both introduced clearly and lands firmly at the end. Why does PREP repeat the Point at the start and end? The opening Point establishes your thesis before the reader has any context — this follows the communication principle of 'tell them what you are going to tell them.' The closing Point restates the thesis after the reader has absorbed the Reason and Example, reinforcing the argument at the moment it is most likely to be retained. Research on persuasion consistently shows that readers are most influenced by the first and last thing they read — PREP places your message at both positions. Is PREP only for formal writing like speeches and essays? No. PREP is equally powerful for business emails, executive briefings, Slack messages, meeting arguments, pitch decks, LinkedIn posts, and any context where you need to make a point clearly and persuasively. Its four-beat structure is compact enough to work in three sentences and structured enough to anchor a 1,000-word argument. What counts as a good Example in PREP? A strong Example in PREP is specific, concrete, and directly illustrates the Reason. In order of persuasive strength: quantitative case studies or data, named real-world examples, analogies from adjacent domains, and illustrative hypothetical scenarios. Weak examples are vague ('many companies have seen success'), anecdotal without context, or only tangentially related to the Reason you are supporting. How is PREP different from the five-paragraph essay structure? The five-paragraph essay has an introduction, three loosely connected body paragraphs, and a conclusion. PREP insists on a single primary reason and a single example rather than three separate threads — this makes PREP arguments tighter and more memorable. For a complex argument with multiple reasons, use multiple PREP cycles rather than expanding a single PREP beyond its natural scope. Can I use PREP for counter-argument structures? Yes. A powerful technique is to use a first PREP cycle to steelman the opposing view and a second PREP cycle to present your own position. This demonstrates intellectual honesty, pre-empts objections, and makes your final Point restatement significantly more persuasive because the reader can see you have considered the alternatives before arriving at your conclusion. How many PREP cycles should I use in a long piece? Each significant argument or sub-point in your piece can be its own PREP cycle. A 1,000-word opinion piece might contain three or four PREP cycles, each addressing a different reason to support the overarching thesis. The opening and closing of the full piece can also follow the PREP structure at a macro level, with the individual argument cycles nested within it. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/prompt-chaining [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Prompt Chaining Workflow Intermediate # Prompt Chaining Break complex tasks into sequential prompts where each output becomes the next input — building reliable, inspectable AI pipelines that outperform any single monolithic prompt. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as LLM Pipelines, Sequential Prompting Used in LangChain, LlamaIndex, custom APIs Key benefit Debuggable, modular, scalable Best for Multi-stage workflows, document processing **TL;DR:** Prompt Chaining decomposes a complex task into a sequence of focused prompts where each output feeds the next, applying a single-responsibility principle to AI workflows for more reliable and debuggable results. Use it for multi-stage tasks — research, drafting, editing, formatting — where a single prompt produces shallow or unreliable output across all stages simultaneously. Map the full task into ordered steps before prompting, then validate each output before passing it to the next chain link. ## What is Prompt Chaining? Prompt Chaining is the practice of decomposing a complex task into a sequence of focused, individual prompts, where the output of each prompt becomes the input for the next. Instead of asking one prompt to do everything, a chain assigns each step exactly one transformation to perform. This mirrors software engineering's single responsibility principle: each function (prompt) does one thing well, and the composition of functions produces the complex behavior. The result is a pipeline that is: - **More reliable:** Each step is easier for the model to do correctly than the entire task at once. - **More debuggable:** You can inspect intermediate outputs to pinpoint exactly where a failure occurred. - **More modifiable:** Change one step without rewriting the entire pipeline. - **Context-window safe:** Large tasks can be broken across multiple calls, each with a fresh context. ## When to Use Prompt Chaining 📄 ### Document Processing Extract → classify → summarize → format: multi-stage document pipelines for contracts, research papers, or support tickets. ✍️ ### Long-Form Content Creation Research → outline → draft section by section → edit → SEO-optimize: each step focused and high-quality. 🔍 ### Research Synthesis Gather sources → extract key claims → identify contradictions → synthesize → write summary: structured information flows. 💻 ### Code Generation Pipelines Spec → architecture → implement → test → document: each stage builds on the structured output of the previous one. 🔄 ### Data Transformation Parse → normalize → enrich → validate → format: ETL-style pipelines for structured data processing. 🧐 ### Multi-Stage Review Draft → fact-check → tone review → final edit: quality gates between each stage with human or automated checkpoints. ## How to Design a Prompt Chain 1. 1 ### Map the task as a flowchart Draw (or write out) each distinct stage of the task. Identify what each step receives as input, what transformation it performs, and what it outputs. Every distinct transformation should be its own chain link. 2. 2 ### Define clean inter-step schemas Specify the exact format each step should output — ideally structured JSON or Markdown with clear headers. Clean, predictable output from step N ensures reliable input for step N+1. Avoid free-form prose between automated chain steps. 3. 3 ### Write each prompt with full context Each prompt should be self-contained: include the relevant output from previous steps, clear instructions for the current step, and the required output format. Don't assume the model "remembers" earlier steps. 4. 4 ### Add validation steps at key junctions After critical steps (especially early in the chain), add a quick validation prompt: "Does this outline meet these criteria? Reply YES or list what's missing." This catches failures before they compound through the rest of the chain. ## Prompt Examples 3-Step Blog Post Chain ``` --- STEP 1: Research & Outline --- You are a technical content strategist. Analyze the following article and create a structured outline for a response post that takes a contrarian perspective. Output as JSON: { "title": "...", "hook": "...", "sections": [{"heading": "...", "key_points": ["..."]}], "cta": "..." } Article: [article text] --- STEP 2: Draft (uses Step 1 output) --- You are a senior technical writer. Write a 900-word blog post following this outline exactly. Maintain a direct, evidence-based, slightly provocative tone. Outline: [JSON from Step 1] --- STEP 3: SEO Edit (uses Step 2 output) --- You are an SEO editor. Optimize this blog post for the keyword "prompt engineering best practices". Requirements: keyword in H1, at least 3 natural uses in body, add a FAQ section with 3 questions, ensure meta description under 160 chars. Draft: [text from Step 2] ``` Customer Feedback Analysis Pipeline ``` --- STEP 1: Extract & Classify --- Extract structured data from this customer feedback. Output JSON: { "sentiment": "positive|negative|neutral", "category": "billing|product|support|other", "key_issue": "one sentence", "urgency": "low|medium|high" } Feedback: "I've been charged twice this month and support has not responded in 5 days. I'm considering canceling." --- STEP 2: Route & Prioritize (uses Step 1) --- Based on this classified feedback, determine: 1. Which team should handle this (Billing / Product / Support) 2. Response SLA (hours) 3. Draft a 2-sentence acknowledgment email Classification: [JSON from Step 1] ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Each step is simpler — higher accuracy per transformation | Multiple API calls — higher latency and cost than single prompts | | Fully debuggable — inspect and fix individual steps | Errors in early steps propagate and amplify downstream | | Modular — swap out individual steps without rebuilding the chain | Requires upfront design and schema definition | | Handles tasks too large for a single context window | More engineering overhead for automated pipelines | ## Prompt Chaining Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Step {{ step_number }} of {{ total_steps }}: {{ step_name }} Input from the previous step: """ {{ previous_output }} """ Your only job in this step: {{ single_responsibility }} Do not {{ out_of_scope }} — later steps handle that. Return the result as {{ handoff_format }} so the next step can consume it directly. ``` ## Prompt Flows That Use Prompt Chaining Prompt Chaining is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The API Designer](https://www.promptedit.app/prompt-flow/the-api-designer) · Software Development ## Frequently Asked Questions What is Prompt Chaining? Prompt Chaining is a technique where you break a complex task into a sequence of smaller, focused prompts, where the output of each prompt becomes the input for the next. Each prompt in the chain handles one specific step, and together they form a pipeline that produces a result more reliably and with higher quality than a single monolithic prompt could achieve. Why use Prompt Chaining instead of one long prompt? A single large prompt forces the model to hold too many instructions in focus simultaneously, which degrades performance on each individual step. Chaining lets each prompt excel at one narrow task. It also makes the pipeline easier to debug (you can inspect intermediate outputs), easier to modify (change one step without rewriting everything), and more resilient (you can retry just the failed step). When should I use Prompt Chaining? Use Prompt Chaining when: the task has multiple distinct stages (research → outline → draft → edit), when intermediate outputs need to be validated or reformatted before the next step, when different steps require different expertise or personas, when the total task would exceed the context window in a single prompt, and when you need to insert human review between automated steps. How do I design a prompt chain? Start by mapping the task as a flowchart: identify each distinct transformation step, define what each step receives as input and produces as output, ensure each output is in a format the next step can consume cleanly. Keep each prompt focused on exactly one transformation. Define clear output schemas (structured JSON works well) when the chain needs to be automated. What is the difference between Prompt Chaining and ReAct? Prompt Chaining is a sequential pipeline designed by the human: each step is predefined, and outputs flow forward in a fixed sequence. ReAct is dynamic: the model itself decides which actions to take at each step based on intermediate observations. Chaining is deterministic and predictable; ReAct is adaptive and flexible. For well-understood workflows, use Chaining; for open-ended research or agent tasks, use ReAct. How do I handle errors in a prompt chain? Build validation steps between chain links: after each major transformation, add a validation prompt that checks the output quality before proceeding. Use structured output schemas (JSON) at each step to catch format errors early. For automated chains, implement retry logic on individual steps rather than rerunning the entire pipeline. Human-in-the-loop checkpoints are valuable for high-stakes chains. Can I use Prompt Chaining with AI APIs programmatically? Yes — Prompt Chaining is the backbone of most production AI pipelines. Frameworks like LangChain, LlamaIndex, and Haystack provide built-in chain abstractions. In raw API implementations, you simply call the API with each prompt, extract the relevant output, format it as the next prompt's input, and repeat. Structured outputs (JSON mode) make this much more reliable. ## Origin & Sources 1. Wu et al. (2022). [AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts](https://arxiv.org/abs/2110.01691) . arXiv / CHI 2022. 2. Anthropic (2026). [Prompt engineering overview](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview) . Claude Docs. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/prompt-refinement-loop [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Prompt Refinement Loop Iterative Technique Intermediate # Prompt Refinement Loop (PRL) Treat prompts like code — draft, test against real inputs, identify failure modes, revise, and repeat. A disciplined loop that turns guesswork into measurable improvement. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Prompt iteration, prompt optimization loop Can be automated Yes, with a meta-prompt evaluator Key tool Structured test set + evaluation rubric Best for Production prompts, quality-critical applications **TL;DR:** The Prompt Refinement Loop is a systematic, test-driven process for improving prompts — maintain a test set, measure outputs against a rubric, categorize failure modes, and make targeted revisions rather than intuition-based edits. Use it for production prompts that must perform reliably across diverse real-world inputs, not just your mental test case. Start by collecting 10-20 representative inputs that cover normal cases and known edge cases before writing your first iteration. ## What is a Prompt Refinement Loop? A Prompt Refinement Loop is a disciplined, iterative process for improving the quality and reliability of AI prompts. Rather than editing prompts based on intuition or the last observed failure, it applies a software-engineering mindset: maintain a test set, measure outputs against a rubric, categorize failure modes, and make targeted revisions that are verifiable against the full test set. The key insight is that prompts are specifications — and like any specification, they need to be tested against real-world inputs to discover the gap between what you intended and what the model actually does. Most prompts that "work" for the developer's mental test case fail silently on edge cases, unusual phrasings, or inputs the developer didn't anticipate. The loop has two modes: - **Manual:** The prompt engineer reviews outputs, identifies failure modes, and revises. Slower but allows nuanced judgment and qualitative assessment. - **Automated:** A meta-prompt evaluator (a second LLM call) scores outputs against a rubric, aggregates failure patterns, and suggests prompt edits. Scales to large test sets and enables CI/CD-style prompt quality gates. ## When to Use the Prompt Refinement Loop 🏭 ### Production Prompt Engineering Any prompt that will run thousands of times in a live application needs systematic refinement before deployment — a single failure mode repeated at scale is costly. 📋 ### Template Development When building reusable prompt templates for teams, the loop ensures the template handles the full range of inputs that different users will provide, not just the template author's use case. 🎯 ### Quality-Critical Applications Legal drafting, medical summarization, financial analysis, and compliance use cases where prompt failures have real consequences require rigorous iterative validation. 🔄 ### Model Migration When switching models (for example moving to a newer GPT, Claude or Gemini generation, or from a large model to a smaller, cheaper one), existing prompts often break in unexpected ways. Running the refinement loop against your test set catches regressions before they reach users. 🧪 ### A/B Prompt Testing Comparing two prompt variants on the same test set with the same rubric gives an objective measure of which performs better — replacing subjective preference with data. 📈 ### Continuous Improvement As user inputs in production reveal new failure modes, feeding those back into the test set and running refinement cycles keeps prompt quality improving over the product lifecycle. ## How to Run a Prompt Refinement Loop 1. 1 ### Draft the initial prompt Write your best first attempt at the prompt. Don't over-engineer it — the loop will surface what's missing. A good starting prompt clearly states the task, the output format, and any critical constraints. 2. 2 ### Build a representative test set Collect 10–30 test inputs that span the expected input distribution: typical cases, edge cases, ambiguous inputs, and inputs that represent known-difficult scenarios. If you have production data, sample from real user queries. Record the expected output or define acceptance criteria for each test case. 3. 3 ### Run the prompt and evaluate outputs Generate outputs for all test inputs. Score each output against your rubric (accuracy, format compliance, tone, completeness, etc.). For automated evaluation, use a meta-prompt evaluator that scores and categorizes each output. Record overall pass rate and scores by dimension. 4. 4 ### Categorize failure modes Group failures by type: under-specification, format drift, scope creep, edge-case collapse, tone inconsistency, or hallucination. Identifying the pattern is more important than fixing individual failures — one prompt change that addresses a failure mode fixes all instances of that category simultaneously. 5. 5 ### Revise the prompt and re-run Make one targeted change that addresses the dominant failure mode. Re-run the full test set. Measure improvement in pass rate and check that the change didn't introduce regressions on previously passing cases. Version your prompt with the pass rate recorded alongside it. 6. 6 ### Repeat until the quality threshold is met Continue cycling until pass rate meets your defined threshold or improvement stalls. When improvement stalls, consider whether the ceiling is a prompt limitation (requiring few-shot examples or chain-of-thought) or a model limitation (requiring a more capable model or fine-tuning). ## Prompt Examples Refinement Cycle — Before and After with Failure Analysis ``` ## Prompt Under Evaluation — Version 1 You are a customer support agent for an e-commerce company. Answer the customer's question helpfully. --- ## Test Input 1 Customer: "My order hasn't arrived and it's been 3 weeks." ## Evaluation Score: 2/5. The prompt provides no context about policies, tone, or response structure. The model invented a refund policy that doesn't match our actual policy. ## Failure Mode Identified Under-specification: no policy context, no tone guidance, no required response elements. ## Revised Prompt — Version 2 You are a customer support agent for ShopFast, a friendly e-commerce company. Our shipping policy: standard delivery takes 5–7 business days; international takes 10–14. Orders over 21 days late qualify for a replacement or full refund. When responding to customers: 1. Acknowledge their concern with empathy (1 sentence) 2. Explain the relevant policy clearly 3. State the specific next step they should take 4. End with an offer to help further Tone: warm, professional, solution-focused. Never blame the customer. ``` Meta-Prompt Evaluator Template ``` ## Meta-Prompt Evaluator You are a prompt quality evaluator. You will be given: - A prompt template - A test input - The model's output for that input Score the output on these criteria (1–5 each): - Accuracy: Does the output correctly address the input? - Format compliance: Does it match the required structure? - Tone: Is the tone appropriate for the use case? - Completeness: Are all required elements present? - Conciseness: Is the output appropriately brief without missing content? After scoring, identify the top failure mode (if any) from this list: [under-specification, format-drift, scope-creep, edge-case-collapse, tone-inconsistency, hallucination] Output as JSON: { "scores": { "accuracy": N, "format": N, "tone": N, "completeness": N, "conciseness": N }, "overall": N, "top_failure_mode": "...", "improvement_suggestion": "One specific change to the prompt that would fix this failure" } --- PROMPT TEMPLATE: [paste prompt here] TEST INPUT: [paste input here] MODEL OUTPUT: [paste output here] ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Produces measurably better prompts with documented improvement history | Requires upfront investment in building a quality test set | | Catches regressions when prompts or models are updated | Manual evaluation is time-intensive for large test sets | | Can be fully automated with a meta-prompt evaluator | Automated evaluators can themselves be unreliable or biased | | Builds institutional knowledge about what failure modes a prompt is susceptible to | Test sets can become stale and no longer represent real-world input distribution | ## Prompt Refinement Loop Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Here is a prompt and an output it produced. Prompt: """ {{ current_prompt }} """ Output: """ {{ output }} """ Evaluate the output against this rubric: {{ rubric }} 1. List each failure and classify it (missing context, ambiguous instruction, wrong format, other). 2. Propose the smallest change to the prompt that fixes each failure. 3. Return the revised prompt in full. ``` ## Prompt Flows That Use Prompt Refinement Loop Prompt Refinement Loop is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The PoC Builder](https://www.promptedit.app/prompt-flow/the-poc-builder) · Software Development ## Frequently Asked Questions What is a Prompt Refinement Loop? A Prompt Refinement Loop is a systematic iterative process for improving prompt quality: draft a prompt, test it on representative inputs, evaluate the outputs against a quality rubric, identify failure modes, revise the prompt to address them, and repeat. It treats prompts like software — subject to versioning, testing, and incremental improvement rather than one-shot guesswork. How is a Prompt Refinement Loop different from just editing a prompt? Ad-hoc editing relies on intuition and tends to fix the last observed failure while introducing new ones. A Prompt Refinement Loop is disciplined: it uses a fixed test set of representative inputs, a defined evaluation rubric, and structured failure mode categorization. Changes are made to address specific, measured failure patterns — not vague impressions — so improvements are verifiable and regressions are caught. What makes a good test set for prompt refinement? A good prompt test set covers the full range of expected inputs: typical cases, edge cases, adversarial inputs (ambiguous phrasing, missing information, conflicting constraints), and cases from known failure categories. Include at least 10–20 examples spanning the distribution of real-world inputs. If you have production logs, sample from actual user queries rather than inventing synthetic test cases. Can the Prompt Refinement Loop be automated? Yes. With a meta-prompt evaluator — a second LLM call that scores outputs against a rubric — the loop can run automatically: generate outputs for all test inputs, score them, aggregate failure modes, and prompt the model to suggest prompt improvements. Tools like PromptFoo, LangSmith, and Braintrust support automated prompt evaluation pipelines that operationalize this loop. How do I know when to stop refining a prompt? Stop when the prompt achieves an acceptable pass rate on your test set (define this threshold upfront — e.g., 90% of outputs rated 4+/5) and the failure modes that remain are either acceptable edge cases or unfixable at the prompt level (requiring fine-tuning or different model selection instead). Also stop if several cycles pass without measurable improvement — this signals a ceiling for prompt-level optimization. What are the most common failure modes found during prompt refinement? The most common failure modes are: (1) under-specification — the prompt is ambiguous and the model makes inconsistent assumptions; (2) format drift — outputs don't match the required structure; (3) scope creep — the model adds unrequested content; (4) edge-case collapse — the prompt works for typical inputs but breaks on unusual ones; (5) tone/style inconsistency — the model interprets style instructions differently across runs. Should I version control my prompts? Absolutely. Prompt version control is essential for production systems. Store prompts in plain text files alongside their test sets and evaluation scores. Use git or a dedicated prompt management tool (LangSmith, Humanloop, PromptLayer) to track changes. When a deployment issue occurs, you need to be able to roll back to a known-good prompt version just as you would with code. ## Origin & Sources 1. Madaan et al. (2023). [Self-Refine: Iterative Refinement with Self-Feedback](https://arxiv.org/abs/2303.17651) . arXiv / NeurIPS 2023. 2. OpenAI (2026). [Prompt engineering guide](https://platform.openai.com/docs/guides/prompt-engineering) . OpenAI Platform Docs. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/quest [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › QUEST Framework Structured Framework Intermediate # QUEST Framework Structure research and information-gathering prompts with five precision components — Question, Understanding, Expectation, Scope, and Time — for focused, deliverable-ready answers. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Question · Understanding · Expectation · Scope · Time Best for Research, analysis, scoped queries Output control High — scope + time constrain depth **TL;DR:** QUEST (Question, Understanding, Expectation, Scope, Time) is a research-focused framework that adds explicit scope and temporal constraints to information-gathering prompts, preventing sprawling, unfocused answers. Use it for competitive research, market analysis, and expert Q&A where a vague query would produce a shallow, generic overview. Use the Scope component to exclude irrelevant angles and the Time component to anchor the analysis to the correct period or deadline. ## What is the QUEST Framework? QUEST is a structured prompt engineering framework designed to bring precision and scope control to research, information-gathering, and analytical tasks. The acronym stands for **Question, Understanding, Expectation, Scope, and Time** — five components that together transform a vague query into a well-specified brief. - **Q — Question:** State the core inquiry - **U — Understanding:** Provide background context - **E — Expectation:** Define the output format - **S — Scope:** Draw the coverage boundaries - **T — Time:** Set a length or depth constraint The framework addresses one of the most common failure modes in AI prompting: the model either misunderstands the background context, produces output in the wrong format, or over-generates content that goes far beyond what was needed. QUEST eliminates each of these problems by making every relevant constraint explicit before the model begins generating. Unlike frameworks focused on creative tasks, QUEST is optimized for situations where accuracy, relevance, and bounded output are critical — making it a natural fit for professional, academic, and technical queries. ## When to Use the QUEST Framework 🔍 ### Research Summaries Gather information on a defined topic with clear boundaries — ideal for competitive research, literature reviews, and market analysis. 📋 ### Board & Executive Reports Produce structured, presentation-ready outputs where length and format are non-negotiable constraints. 🏗️ ### Technical Scoping Ask architectural or engineering questions while explicitly excluding out-of-scope concerns that would otherwise dilute the answer. 📚 ### Academic Analysis Define a precise research question, provide your existing understanding, and constrain the scope to a specific domain or time period. 🧩 ### Problem Diagnosis Frame a troubleshooting question with full background context and a clear expectation of the diagnostic format the answer should follow. ⏱️ ### Time-Boxed Deliverables When you need an answer that fits a specific slot — a 2-minute brief, a one-page summary — the Time component enforces the right depth. ## How to Use the QUEST Framework 1. 1 ### Question — State the core inquiry Begin with a single, clear question that captures exactly what you want to know. Avoid bundling multiple questions. If you have sub-questions, list them beneath the primary question. A sharp Question component prevents the model from guessing what you actually want. 2. 2 ### Understanding — Provide background context Brief the model as you would a human expert. Share your role, your current knowledge level, what you have already tried, and any domain-specific definitions. This prevents the model from over-explaining basics or making wrong assumptions about your situation. 3. 3 ### Expectation — Define the output format Specify exactly what the response should look like: bullet points, a numbered list, a report with sections, a comparison table, or a single paragraph. Include any formatting requirements, citation preferences, or structural constraints. 4. 4 ### Scope — Draw the coverage boundaries State explicitly what is in and out of scope. Use language like "Cover only...", "Do not address...", and "Limit to...". Scope is your primary tool for preventing irrelevant tangents and keeping the response tightly relevant. 5. 5 ### Time — Set a length or depth constraint Specify the expected depth of the response. This might be a word count, a reading-time target, or a qualitative signal like "a high-level overview" or "an exhaustive analysis". Time constraints help calibrate how much effort the model should invest in each point. ## Prompt Examples HR Research — Employee Attrition Analysis ``` Question: What are the main factors driving employee attrition in tech companies in 2024? Understanding: I am an HR manager at a 200-person SaaS company. We have seen a 22% annual attrition rate over the past year — higher than the industry average. I am preparing a board presentation and need credible data points and root causes, not generic advice. Expectation: A structured report with 4-6 identified factors, each with a brief explanation and one supporting data point or study where available. Use bullet points under each factor. Scope: Focus on mid-size tech companies (50–500 employees) in North America. Do not cover macro-economic conditions or factors unrelated to the workplace environment. Time: The response should be readable in under 5 minutes — approximately 400–500 words. ``` Technical — Multi-Tenant API Architecture ``` Question: How should I structure a REST API for a multi-tenant SaaS application? Understanding: I am a backend developer building a B2B product with approximately 50 tenants at launch, expecting to scale to 500 within 18 months. I am using Node.js and PostgreSQL. I have basic familiarity with REST but have not built a multi-tenant system before. Expectation: A recommended architecture with the key design decisions explained. Include a short comparison of row-level tenancy vs. schema-per-tenant vs. database-per-tenant, and state your recommended approach with a rationale. Scope: Focus on data isolation and API design. Do not cover authentication providers, billing integration, or frontend architecture. Time: Provide a concise technical overview — aim for 500–600 words with a clear recommendation at the end. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Eliminates ambiguity through explicit scoping and context | More overhead than zero-shot prompting for simple questions | | Time component prevents over-generation and verbose responses | Rigid scoping can suppress useful context the model might otherwise offer | | Highly adaptable to professional and technical domains | Less suited for open-ended creative or generative tasks | | Works well with both simple and complex research queries | | ## QUEST Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Question: {{ question }} Understanding: {{ background }} Expectation: {{ output_format }} Scope: {{ in_and_out_of_scope }} Time: {{ length_or_depth }} ``` ## Prompt Flows That Use QUEST Framework QUEST Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Research Synthesizer](https://www.promptedit.app/prompt-flow/the-research-synthesizer) · Analysis & Research ## Frequently Asked Questions What does QUEST stand for in prompt engineering? QUEST is an acronym for Question, Understanding, Expectation, Scope, and Time. Each component adds a distinct layer to your prompt: the core question defines what you're asking, understanding provides the background context, expectation sets the output format, scope draws boundaries around the coverage, and time adds a deadline or length constraint. How is QUEST different from other structured frameworks like RISEN or CRISPE? QUEST is uniquely focused on information-gathering and research tasks. Where RISEN emphasizes role and steps, and CRISPE adds a persona, QUEST's Time component enforces a constraint on depth or length — making it ideal when you need a scoped, deliverable-ready answer rather than an open-ended exploration. Do I need to include all five QUEST components every time? No. For simple queries, Question + Expectation alone may suffice. The Scope and Time components become critical when you need to prevent the model from over-generating or going off-topic. Understanding is most valuable when the model might make incorrect background assumptions. Start with the components that address your biggest failure mode. What should I put in the 'Understanding' component? Understanding captures the context and background knowledge the model needs before answering. Include your level of expertise, relevant constraints, any domain-specific definitions, or prior attempts at solving the problem. Think of it as briefing a consultant before they start work. How do I write an effective Scope component? Define what is explicitly in and out of coverage. For example: 'Cover only publicly available data. Exclude financial projections and speculative analysis.' Scope is most powerful when the question is broad by nature — it prevents irrelevant tangents and keeps the response focused. What is the Time component used for? Time serves a dual purpose: it can mean a real-world deadline ('summarize findings by end of day'), a length constraint ('respond in under 300 words'), or a depth limit ('a quick overview, not an exhaustive analysis'). It signals urgency and expected effort level to the model. Can QUEST be used for creative writing tasks? QUEST is best suited to research, analysis, and information retrieval tasks. For creative writing, the IDEA framework (Intent, Details, Examples, Adjustments) is a better fit because it accommodates iteration and stylistic examples. QUEST's rigid scoping can suppress creative exploration. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/race [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RACE Framework Structured Framework Beginner # RACE Framework Four memorable components — Role, Action, Context, Expectation — that turn vague requests into precise, high-quality AI prompts in under two minutes. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Role — Action — Context — Expectation Best for Everyday writing, analysis, and Q&A tasks Complexity Low — easy to memorize and apply Token cost Low **TL;DR:** RACE (Role, Action, Context, Expectation) is a versatile four-component framework that covers the most impactful prompt variables — expert persona, specific task, background context, and output format — in a memorable, learnable structure. Use it as a general-purpose template for professional writing, analysis, and content generation tasks of any complexity. RACE is the ideal framework to standardize across a team because its four components are non-overlapping and intuitive for non-technical users. ## What is the RACE Framework? The RACE framework is a four-component prompt structure that helps you write clear, complete, and effective prompts for any AI model. Each letter stands for a distinct ingredient: **Role** (who the AI should be), **Action** (what it should do), **Context** (the background information it needs), and **Expectation** (what the output should look like). - **R — Role:** Define the expert persona - **A — Action:** State exactly what to do - **C — Context:** Supply the background - **E — Expectation:** Specify the output What makes RACE so popular is its balance of simplicity and completeness. Unlike bare-bones structures such as RTF, RACE includes a dedicated Context slot, which is often the missing piece that causes AI responses to feel generic. Unlike heavier frameworks such as RISEN or CRISPE, RACE does not require you to supply worked examples or fine-grained stylistic directives — making it fast to compose for everyday tasks. RACE works across a wide range of use cases: content writing, code review, data analysis, customer support drafts, research summaries, and more. Think of it as a universal prompt template you can internalize and apply in seconds. ## When to Use the RACE Framework ✍️ ### Content Writing Assign a writer persona, specify the content type and key messages, provide audience context, and set tone and word-count expectations for on-brand copy every time. 📧 ### Professional Emails Set the Role to a communications professional, Action to drafting or rewriting an email, Context to the relationship and purpose, and Expectation to length and formality level. 🔍 ### Research Summaries Give the AI an expert analyst role, ask it to summarize a topic, supply the source material or domain constraints in Context, and specify the format (bullet points, executive brief, etc.). 💻 ### Code Review & Generation Assign a senior developer role, specify the code action needed, provide language and framework context, and set expectations around comments, style guides, and explanation depth. 🎓 ### Educational Explanations Set the AI as a teacher, ask it to explain a concept, describe the learner's level in Context, and specify whether you want analogies, examples, or a structured lesson format. 📊 ### Data Analysis Narratives Assign a data analyst role, ask for interpretation of findings, provide the dataset or summary in Context, and expect a written narrative with key takeaways for a non-technical audience. ## How to Use the RACE Framework 1. 1 ### Role — Define the expert persona Start by telling the AI who it should be. Be specific: "You are a senior UX designer with 10 years of mobile experience" gives far better results than "You are a designer." Include the domain, seniority, and any relevant specialization. A well-crafted Role primes the model to draw on the right vocabulary, assumptions, and problem-solving approach. 2. 2 ### Action — State exactly what to do Use a clear, active verb to describe the task: write, summarize, compare, debug, translate, rewrite, explain, draft, generate. Follow the verb with the specific object. Avoid vague language like "help me with" or "talk about." The more precisely you define the action, the less the model has to guess. 3. 3 ### Context — Supply the background Context is the component most often skipped and most often responsible for generic outputs. Include who the audience is, what constraints apply, what the existing situation is, and any domain-specific facts the AI needs to know. Good context transforms a generic answer into a tailored, situationally aware response. 4. 4 ### Expectation — Specify the output Tell the AI exactly what a successful response looks like. Cover format (numbered list, markdown table, prose paragraphs), length (word count, number of items), tone (formal, conversational, technical), and any hard constraints (no jargon, include a call to action, stay under 150 words). This is the component that controls output quality most directly. ## Prompt Examples Financial Explanation for Beginners ``` Role: You are a senior financial advisor with 20 years of experience in personal finance and investment planning. Action: Write a plain-language explanation of how compound interest works and why starting early matters. Context: My audience is recent college graduates who have never invested before and are intimidated by financial jargon. They have roughly $200–$500/month to invest. Expectation: Produce a 300-word explanation with one concrete numerical example showing the difference between starting at 22 versus 32. Use encouraging, jargon-free language. ``` UX Writing — Error Message Rewrite ``` Role: You are an experienced UX writer specializing in mobile app onboarding. Action: Rewrite the following error message to be clearer and friendlier: "Error 403: Access denied. Insufficient privileges." Context: This message appears in a consumer fitness app when a free-tier user tries to access a premium feature. The app's brand voice is warm, motivating, and conversational. Expectation: Provide three alternative error message options, each under 20 words, that explain what happened and nudge the user toward upgrading — without feeling pushy. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Easy to memorize — four intuitive components | No dedicated slot for examples — add them manually when needed | | Covers all the essentials without unnecessary complexity | Less precise than RISEN or CRISPE for highly nuanced tasks | | Works across almost any task type or domain | Expectation component requires some practice to write well | | Significantly better outputs than unstructured prompts | | ## RACE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: You are {{ role }}. Action: {{ action }} Context: {{ context }} Expectation: {{ expected_output }} ``` ## Frequently Asked Questions What does RACE stand for in prompt engineering? RACE stands for Role, Action, Context, and Expectation. You define who the AI should be (Role), what it should do (Action), the background information it needs (Context), and what the output should look like in terms of format and quality (Expectation). Together these four components create a complete, unambiguous prompt. How is RACE different from RISEN or CRISPE? RACE is simpler and faster to compose than RISEN (which adds Instructions and Narrowing) or CRISPE (which adds Style and Examples). RACE is ideal for everyday tasks where you need a structured prompt quickly. RISEN and CRISPE are better when you need fine-grained output control or want to supply worked examples. Do I always need all four RACE components? Not always. For very simple requests, you might omit Role or abbreviate Context. However, including all four consistently produces better results, especially Expectation — this is the component most people skip, and it's the one that most directly controls output quality and format. What goes in the Expectation component of RACE? Expectation covers the desired output format (bullet list, essay, table, code block), the length or scope (200 words, three options, one paragraph), the quality bar (professional tone, beginner-friendly, peer-reviewed standard), and any constraints on what the response should or should not include. Can I use RACE for coding tasks? Yes. Set Role to the type of developer or engineer you need, Action to the specific coding task, Context to the codebase constraints and language/framework, and Expectation to details like code style, whether to include comments, and whether to explain the code alongside it. How does RACE compare to RTF (Role, Task, Format)? RTF is a minimal three-part framework where Format covers output structure. RACE expands this by splitting task guidance into Action (verb-driven) and adding a dedicated Context component for background information. RACE is more thorough and produces more tailored results than RTF for complex tasks. Is RACE suitable for beginners to prompt engineering? RACE is one of the most beginner-friendly structured frameworks. The four letters are easy to remember, each component has a clear and intuitive purpose, and you can build your first RACE prompt in under two minutes. It serves as an excellent gateway to more advanced frameworks like RISEN or CRISPE. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/rain [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RAIN Framework Structured Framework Intermediate # RAIN Framework Produce measurable, bounded AI outputs by combining a clear Role, Aim, working Input, and a Numeric Target — the framework that eliminates vague, open-ended responses. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Stands for Role · Aim · Input · Numeric Target Signature feature Numeric Target for measurable output Best for Data tasks, analytics, KPI-driven work Complexity Medium — 4 focused components **TL;DR:** RAIN (Role, Aim, Input, Numeric Target) is a four-component framework whose Numeric Target component forces a specific, measurable output constraint — word counts, item counts, percentages, scores — preventing the AI from producing open-ended responses when bounded output is required. Use it for deliverables with strict size or quantity requirements: top-N lists, character-limited copy, fixed-length summaries. Specify the Numeric Target as a hard limit, not a suggestion. ## What is the RAIN Framework? RAIN is a prompt engineering framework built around four components: **Role**, **Aim**, **Input**, and **Numeric Target**. Its defining innovation is the Numeric Target — a specific, quantifiable output constraint that forces the AI to produce a bounded, measurable response rather than an open-ended essay. - **R — Role:** Assign the Expert Identity - **A — Aim:** State the Overall Goal - **I — Input:** Provide the Working Material - **N — Numeric Target:** Set the Measurable Constraint Most prompt frameworks tell the AI what to do but leave "how much" to the model's discretion. RAIN closes that gap. Whether you need exactly 10 ideas, a 100-word summary, the top 5 root causes, or a score out of 10 for each option, the Numeric Target makes the success criterion explicit before the AI begins. The framework is especially well-suited for data analysis, content generation at scale, diagnostic tasks, and any situation where the output must meet a quantifiable standard to be useful. ## When to Use the RAIN Framework 📊 ### Data Analysis Feed the AI a dataset or metrics and set a Numeric Target for how many insights, root causes, or recommendations to surface — preventing bloated, unfocused outputs. 💡 ### Ideation at Scale Generate a precise number of blog titles, product names, ad copy variants, or feature ideas — no more, no less — so you get exactly what you need for a review session. 📝 ### Summarization Summarize long documents, reports, or transcripts to a specific word count or number of bullet points, ensuring output fits your use case without trimming it yourself. 🎯 ### KPI-Driven Recommendations When the task is to improve a metric, RAIN's Numeric Target can specify both the goal (e.g., "improve by 15%") and the number of recommendations to generate. 🔍 ### Competitive Research Analyse competitor data and extract a fixed number of differentiators, threats, or opportunities — keeping research deliverables consistently scoped across projects. 📋 ### Audit & Review Review content, code, or processes and return the top N issues ranked by severity, giving reviewers a prioritised, action-ready list rather than an exhaustive inventory. ## How to Use the RAIN Framework 1. R ### Role — Assign the Expert Identity Start by telling the AI who it is for this task. Choose a role that aligns with the expertise your task requires: "You are a senior data analyst", "You are a B2B content strategist with 10 years of experience". A well-chosen Role primes the model to use domain-appropriate knowledge, vocabulary, and judgment. 2. A ### Aim — State the Overall Goal Describe what you are ultimately trying to achieve in one or two sentences. The Aim is the strategic objective that motivates the task — it gives the AI the "why" so it can make better judgment calls about what to prioritise within the scope of the work. 3. I ### Input — Provide the Working Material Supply the data, content, metrics, or documents the AI needs to work with. This is distinct from background context — Input is the raw material to be processed, analysed, or transformed. Be specific: paste the actual data, link to the source, or describe the inputs precisely so the AI has everything it needs. 4. N ### Numeric Target — Set the Measurable Constraint Define a specific, quantifiable output boundary. Examples: "generate exactly 12 ideas", "summarize in 75 words or fewer", "rank the top 5 options", "score each item out of 10", "list actions for the next 14 days". The Numeric Target is what separates RAIN from vague open-ended prompts — it makes success verifiable at a glance. ## Prompt Examples Content Ideation with Numeric Target ``` Role: You are a senior content strategist with 10 years of experience in B2B SaaS marketing. Aim: Generate high-quality blog post ideas that will drive organic search traffic for a project management software company targeting mid-market teams. Input: Our target audience is operations managers at companies with 50–500 employees. Top pain points: missed deadlines, unclear task ownership, and too many status meetings. Current top-performing content: "How to Run a Weekly Team Standup" (8,400 monthly visits). Numeric Target: Generate exactly 15 blog post ideas. For each idea include: the headline, target keyword, estimated search intent (informational/commercial), and one-sentence content angle. Rank them from highest to lowest estimated traffic potential. ``` Conversion Rate Diagnosis ``` Role: You are a data analyst specialising in e-commerce performance metrics. Aim: Diagnose why our checkout conversion rate dropped from 3.2% to 1.9% over the last 30 days and recommend prioritised fixes. Input: Key data points from the last 30 days: - Mobile checkout abandonment increased from 58% to 74% - Average page load time on checkout increased from 1.8s to 4.1s - A new one-page checkout design was deployed on March 1 - Cart abandonment email open rate: 34% (unchanged) - Top exit page: payment details step (new — previously shipping step) - Browser breakdown: Chrome 61%, Safari 24%, Firefox 15% Numeric Target: Identify the top 5 root causes ranked by likely impact. For each cause provide: a one-sentence diagnosis, supporting data point, and a specific fix with an estimated implementation effort (Low / Medium / High). ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Numeric Target eliminates vague, open-ended responses | Numeric Target can feel artificial for exploratory or open-ended tasks | | Input component ensures the AI has the right working material | Requires clear upfront thinking about what "enough" looks like | | Outputs are immediately verifiable against the stated target | Less suited for nuanced tone-sensitive writing where quantity is irrelevant | | Highly effective for data tasks, ideation, and KPI-driven work | | ## RAIN Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: You are {{ role }}. Aim: {{ aim }} Input: {{ working_material }} Numeric target: {{ measurable_constraint }} ``` ## Frequently Asked Questions What does RAIN stand for? RAIN stands for Role, Aim, Input, and Numeric Target. Role sets the expert identity, Aim defines the overall goal, Input provides the data or content the AI needs to work with, and Numeric Target sets a specific, quantifiable output constraint such as 'generate 10 ideas' or 'summarize in 50 words'. What makes RAIN different from other structured frameworks? The Numeric Target component is RAIN's defining feature. Most frameworks describe what they want but leave output scope vague. RAIN forces you to define a measurable success criterion upfront — a specific count, word limit, score threshold, or other quantifiable boundary — which dramatically reduces overly long or unfocused outputs. When should I use the RAIN framework? Use RAIN for data-heavy, measurable, or analytically-driven tasks: generating a specific number of ideas, summarizing content to a word count, ranking options by score, or diagnosing a fixed number of root causes. It is also excellent for KPI-driven work where the output must meet a quantifiable standard. What counts as a good Numeric Target? A Numeric Target is any measurable output constraint: a count ('generate 10 headlines'), a length ('summarize in 100 words or fewer'), a ranking depth ('identify the top 5'), a score ('rate each option out of 10'), or a time boundary ('list actions for the next 7 days'). The key is that the constraint is specific and verifiable. Can I use RAIN without a true numeric target? Yes, but the framework loses its primary advantage. If your task genuinely has no measurable output constraint, consider whether a different framework like RISEN or APE is a better fit. Alternatively, add a soft constraint such as 'be concise' or 'limit your response to the three most important points' to preserve bounded output. How does the Input component differ from Context in other frameworks? RAIN's Input component is specifically for data, content, or material the AI must process or work with — a dataset, an article to summarize, specific metrics, or a document to analyse. It is more focused than the broad 'Context' component in frameworks like COAST or CARE, which covers background environment rather than raw working material. Does RAIN work well for creative tasks? RAIN works for creative tasks that have measurable outputs — 'generate 20 subject line variations', 'write 5 tagline options', 'produce 3 alternative product descriptions'. For open-ended creative work without natural output boundaries, frameworks that emphasise style and tone (like DEEP or PIVO) may produce better results. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/rasce [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RASCE Framework Structured Prompting Intermediate # RASCE Framework Five precisely targeted components — Role, Audience, Style, Constraints, Examples — that ensure every piece of AI-generated content is perfectly matched to its reader, brand voice, and creative boundaries. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Role · Audience · Style · Constraints · Examples Best for Content marketing, emails, brand writing, audience-specific copy Complexity Medium — five components, two of which require audience research Token cost Medium **TL;DR:** RASCE (Role, Audience, Style, Constraints, Examples) is a five-component framework purpose-built for content tasks where audience fit and stylistic precision are non-negotiable — the dedicated Audience and Style components do more work than the Context field in generic frameworks. Use it for marketing copy, editorial content, and brand communication where "write in a professional tone" produces insufficient results. The Examples component replaces vague adjectives with a concrete style anchor: point to a specific author, publication, or existing piece. ## What is the RASCE Framework? RASCE is a five-component prompt engineering framework built for content and communications tasks where audience fit and stylistic precision matter most. The acronym stands for **Role** (the expert identity the AI should adopt), **Audience** (who will read or use the output), **Style** (the voice, tone, and register), **Constraints** (hard rules the output must respect), and **Examples** (reference material that calibrates tone and quality). What sets RASCE apart from general-purpose frameworks like RACE or RTF is its dedicated Audience and Style components. Most prompting frameworks treat audience as a footnote inside a broader Context field, but for content tasks — marketing copy, onboarding emails, educational materials, internal communications — knowing exactly who you are writing for and how it should sound is the difference between generic and genuinely effective output. The Examples component in RASCE functions as a style anchor rather than a task demonstration. By pointing the AI at a specific publication, author, or brand voice, you replace imprecise adjectives ("make it professional") with a concrete, learnable reference. This produces more consistent stylistic output, especially when the prompt will be reused as a template across multiple pieces of content. ## When to Use the RASCE Framework ✍️ ### Brand Content Writing Maintain a consistent brand voice across blog posts, social copy, and landing pages by anchoring Style and Examples to your brand guidelines and existing high-performing content. 📧 ### Email Campaigns Tailor onboarding sequences, newsletters, and promotional emails precisely to subscriber segments — specifying their seniority, motivations, and knowledge level in the Audience component. 🎓 ### Educational Materials Write lesson content, explainers, or course copy calibrated to a learner's exact level — beginner, intermediate, expert — with a style reference that matches the desired pedagogical approach. 🎯 ### Audience-Specific Marketing Create messaging for distinct buyer personas — technical buyers, executive sponsors, end users — each with their own language, concerns, and decision criteria specified in Audience. 📝 ### Constrained Creative Tasks When word limits, legal constraints, regulatory language, or platform character limits apply, the Constraints component ensures the AI never produces output that violates these boundaries. 🤝 ### Internal Communications Draft memos, announcements, and policy updates that match the formality level and cultural norms of a specific organization or department, grounded in style references from existing internal documents. ## How to Use the RASCE Framework 1. 1 ### Role — Establish the expert identity Define who the AI is in terms of domain expertise, seniority, and relevant experience. The more specific the role, the more calibrated the output. "You are a senior B2B content strategist with a decade of SaaS marketing experience" produces sharper, more informed content than "You are a writer." Include specializations that are directly relevant to the task at hand. 2. 2 ### Audience — Portrait the reader in detail Describe the intended reader as concretely as possible: job title, seniority, industry, technical literacy, motivations, fears, and situational context. What do they already know? What are they trying to accomplish? What vocabulary do they use? This component is where RASCE delivers its greatest advantage — a precise audience portrait eliminates generic, one-size-fits-all output. 3. 3 ### Style — Define voice, tone, and register Specify how the content should feel: formal or conversational, authoritative or warm, dense with data or light and motivating. Go beyond adjectives when you can — reference a specific publication, author, or piece of writing whose voice matches what you want. Describe what to avoid as well as what to achieve. Negative style guidance ("avoid passive voice, avoid corporate jargon") is often as valuable as positive direction. 4. 4 ### Constraints + Examples — Set the rules and the benchmark List hard constraints: word count, mandatory inclusions, prohibited content, structural requirements, legal language, formatting rules. Then supply one or more Examples — a reference piece, a style guide excerpt, or a pasted sample — that shows the AI what great looks like. Constraints tell the model what it must not do; Examples show it what it should aspire to. ## Prompt Examples B2B Blog Post — SaaS Marketing Audience ``` Role: You are a senior B2B content strategist with expertise in SaaS marketing. Audience: Mid-level marketing managers at companies with 50–500 employees who are evaluating project management software. They are data-literate, time-pressed, and skeptical of vendor claims. Style: Authoritative but approachable. Use concrete data points where possible. Avoid hype language, passive voice, and buzzwords like "game-changer" or "seamless." Constraints: The piece must be 600–800 words, include one real-world scenario illustrating ROI, and end with a single clear call to action. Do not mention competitors by name. Examples: Model the tone after Harvard Business Review's practical guides — dense with insight but written in plain English with short paragraphs. Task: Write a blog post titled "Five Signs Your Team Has Outgrown Spreadsheet-Based Project Tracking." ``` Day 1 Welcome Email — Freelancer Onboarding ``` Role: You are an empathetic customer success writer who specializes in SaaS onboarding communications. Audience: First-time users of a time-tracking app who signed up in the last 48 hours. They are freelancers and solo consultants aged 25–45. Many are non-technical and signed up after a frustration with manual invoicing. Style: Warm, encouraging, and brief. Use second-person ("you"), active verbs, and simple vocabulary. The tone should feel like a helpful message from a knowledgeable friend, not a corporate announcement. Constraints: Maximum 150 words. Must include: a welcome line, one key action to complete today, and a low-pressure invitation to reply with questions. No bullet lists — flowing prose only. Examples: Match the brevity and warmth of Notion's onboarding emails — personal, helpful, zero jargon. Task: Write a Day 1 welcome email for new users of the app. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dedicated Audience component eliminates generic, untailored output | Five components take longer to compose than simpler frameworks | | Style plus Examples anchors tone far more reliably than adjectives alone | Audience research is required — lazy audience descriptions reduce quality | | Constraints prevent common output failures before they happen | Less suited to non-content tasks like data analysis or coding | | Excellent as a reusable template for recurring content types | Examples must be chosen carefully — a weak reference produces weak calibration | | Works across any content domain: marketing, education, internal comms | | ## RASCE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: {{ role }} Audience: {{ audience }} Style: {{ style }} Constraints: {{ constraints }} Examples: {{ examples }} ``` ## Frequently Asked Questions What does RASCE stand for? RASCE stands for Role, Audience, Style, Constraints, and Examples. It is a five-component prompting framework designed to produce content that is precisely tailored to a specific reader, written in the right voice, bounded by clear rules, and grounded in a reference tone or format. How is RASCE different from RACE or RISEN? RASCE places dedicated emphasis on Audience and Style — two components that general frameworks like RACE fold into Context or Expectation. This makes RASCE especially well-suited for content and communications tasks where knowing who you are writing for and how it should sound matters as much as what the task is. RISEN focuses more on narrowing scope and adding worked instructions, while RASCE uses Examples to calibrate tone. When should I use RASCE instead of a simpler framework? Use RASCE when the audience is specific and the tone matters significantly — for example, marketing copy, onboarding emails, educational materials, or brand communications. If you are doing a quick one-off task like summarizing a document, a simpler structure like RTF or RACE is faster. RASCE pays dividends when you will reuse the prompt template repeatedly for the same audience. What goes in the Examples component of RASCE? Examples in RASCE are tone and style references, not worked task examples. Point the AI at a publication, brand, author, or specific piece of writing whose voice you want to match. You can also paste a short excerpt directly into the prompt. This reference anchors the model's stylistic output far more reliably than adjectives alone — 'write in a professional tone' is far less effective than 'match the style of The Economist's briefings'. Can I use RASCE for non-written tasks like code generation? RASCE is primarily designed for content and communication tasks, but it can be adapted for code generation. In that context, Audience becomes the intended developer or user of the code, Style maps to code conventions and documentation standards, and Constraints covers language version, performance requirements, and library restrictions. Examples would reference a code style guide or an existing snippet. Do I need to include all five RASCE components every time? Including all five consistently produces the best results, but you can omit Examples when you genuinely have no reference to offer. Audience and Style are the most important components to keep — they are what differentiate RASCE from other frameworks. Dropping Role or Constraints weakens the prompt significantly. As a minimum, aim for Role, Audience, Style, and Constraints. How specific should the Audience description be? As specific as you can make it without inventing fictional details. Include demographics if relevant (age range, job title, seniority), psychographics (motivations, fears, knowledge level), and situational context — why they are reading this, what they already know, what they are trying to accomplish. The more precisely the AI understands the reader, the more tailored the output. Generic audience descriptions produce generic content. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/react-prompting [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › ReAct Prompting Agent Intermediate # ReAct Prompting Interleave explicit reasoning with concrete actions — Thought → Action → Observation — enabling AI models to solve tasks requiring tool use, information retrieval, and dynamic decision-making. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Yao et al., Princeton & Google (2022) Pattern Thought → Action → Observation (repeat) Requires Tool integrations or APIs Used in AI agent frameworks (LangChain, etc.) **TL;DR:** ReAct (Reasoning and Acting) interleaves explicit reasoning steps with tool calls — Thought, Action, Observation — so the model can ground its reasoning in real retrieved information rather than relying solely on training data. Use it for agentic tasks requiring web search, code execution, API calls, or any workflow where current or external information determines correctness. Define available tools explicitly in the system prompt and show a worked Thought-Action-Observation example to set the expected format. ## What is ReAct Prompting? ReAct (Reasoning and Acting) is a prompting paradigm introduced by Yao et al. in 2022 that synergizes two capabilities that were traditionally separate: reasoning (Chain-of-Thought) and acting (tool use). The model follows a structured cycle: - **Thought:** The model reasons about what it knows and what it needs to do next. - **Action:** The model calls an external tool (search, calculator, code executor, API). - **Observation:** The model receives and processes the tool's output. This cycle repeats until the model has enough information to produce a final answer. ReAct is the conceptual backbone of most modern AI agent architectures, including LangChain's agent module, OpenAI's function calling, and Anthropic's tool use API. The key insight is that interleaving reasoning with real observations makes the model's actions more purposeful (it thinks before acting) and its reasoning more grounded (it updates based on real results, not hallucinated facts). ## When to Use ReAct Prompting 🔍 ### Research Tasks Multi-step information gathering where the model needs to search, synthesize, and verify facts from multiple sources before answering. 🧮 ### Calculations & Data Analysis Tasks requiring numeric computation, where the model should call a calculator or code interpreter rather than hallucinate arithmetic. 🤖 ### Autonomous AI Agents Agents that must take actions in systems (send email, update CRM, query database) based on reasoning about their state and goals. 🛒 ### E-commerce & Product Research Price comparison, product lookup, and recommendation tasks that require live data retrieval combined with analytical reasoning. 🗃️ ### Database Interaction Query generation, result interpretation, and iterative refinement of database queries based on intermediate results. 📅 ### Task Orchestration Complex workflows where the next step depends on the result of the previous step — calendar scheduling, resource allocation, multi-system coordination. ## How to Implement ReAct 1. 1 ### Define available tools List the actions the model can take and describe each tool: its name, what it does, its input format, and what it returns. The model needs this to reason about which tool to use. 2. 2 ### Provide a Thought/Action/Observation example Show the model the exact format to follow with one worked example. This establishes the structural pattern before the model begins the actual task. 3. 3 ### Present the task and let the model begin The model generates a Thought and an Action. Your code (or you manually) executes the action, captures the result, and appends it as an Observation. Feed this back to the model. 4. 4 ### Repeat until Final Answer Continue the Thought → Action → Observation cycle until the model generates a "Final Answer" response. Add stopping conditions (max iterations, confidence threshold) to prevent infinite loops. ## Prompt Examples ReAct Trace Format ``` You have access to these tools: - search(query): Search the web and return relevant excerpts - calculate(expression): Evaluate a mathematical expression - lookup(entity): Get a Wikipedia summary of an entity Use the Thought/Action/Observation format. When done, write "Final Answer: ..." Question: What is the population of the capital of France, and how does that compare to the population of Berlin? Thought: I need to find the capital of France and its population, then find Berlin's population to compare. Action: lookup("Paris") Observation: Paris is the capital of France with a population of approximately 2.1 million in the city proper (11M metro). Thought: Now I need Berlin's population. Action: lookup("Berlin") Observation: Berlin is Germany's capital with ~3.7M city population. Thought: I have both figures. Paris city: 2.1M, Berlin: 3.7M. Berlin is larger by ~1.6M people (76% more than Paris city proper). Final Answer: Paris has ~2.1M residents vs. Berlin's ~3.7M. Berlin is approximately 76% more populous than Paris (city proper). ``` Single-Prompt ReAct Simulation ``` Simulate a ReAct agent solving this research task step by step. For each step, show: Thought: [your reasoning] Action: [what tool you'd call and with what input] Observation: [what result you'd expect] Continue until you reach a Final Answer. Task: Find the current market leader in the vector database space, identify their main differentiator, and estimate their market share relative to the top 3 competitors. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Grounds reasoning in real, external data — reduces hallucination | Requires tool infrastructure — not usable without integrations | | Enables truly autonomous agent behavior | More complex to implement than pure prompting techniques | | Reasoning traces make agent behavior interpretable and debuggable | Multiple model calls increase latency and cost | | Foundation of production AI agent frameworks | Poor tool design leads to poor agent behavior | ## ReAct Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Answer the question below. You can use these tools: {{ tool_list }}. Use this loop until you can answer: Thought: reason about what you know and what you still need Action: tool_name[input] Observation: (the tool result will appear here) When you have enough information, write: Final Answer: {{ answer_format }} Question: {{ question }} ``` ## Prompt Flows That Use ReAct Prompting ReAct Prompting is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The PoC Builder](https://www.promptedit.app/prompt-flow/the-poc-builder) · Software Development ## Frequently Asked Questions What is ReAct Prompting? ReAct (Reasoning + Acting) is a prompting framework where the model interleaves reasoning traces with concrete actions in a cyclical pattern: Thought → Action → Observation → Thought → Action → Observation... The model explicitly reasons about what to do next (Thought), takes an action like a search or calculation (Action), and then processes the result (Observation) before reasoning again. This cycle continues until the task is complete. What does ReAct stand for? ReAct stands for Reasoning and Acting. It was introduced by Yao et al. in a 2022 paper and demonstrates that synergizing reasoning traces with task-specific actions improves both performance and interpretability on tasks requiring tool use, information retrieval, and decision-making. When should I use ReAct Prompting? ReAct is ideal when the task requires gathering external information (web search, database queries), performing computations (calculators, code interpreters), taking actions in an environment (API calls, file operations), or making sequences of decisions based on intermediate results. It is the foundation of most modern AI agent architectures. What is the difference between ReAct and Chain-of-Thought? Chain-of-Thought produces reasoning in a single forward pass — the model thinks and then answers without external input. ReAct interleaves reasoning with actual actions and real observations, allowing the model to gather new information mid-task. CoT is for self-contained reasoning; ReAct is for tasks requiring interaction with the world. How do I implement ReAct without an agent framework? In a single prompt or conversation, you can simulate ReAct by explicitly formatting your prompt with Thought/Action/Observation headers. Ask the model to generate a Thought and Action, then you (or your code) provide the Observation, and the cycle continues. Many developers use LangChain, LlamaIndex, or custom orchestration code to automate this loop. What actions can a ReAct agent take? Actions depend on the tools available: web search (Google, Bing API), Wikipedia lookup, code execution (Python interpreter), calculator, database queries, file read/write, API calls (weather, stock prices, CRM systems), and any custom tool you define. The power of ReAct comes from the agent's ability to choose the right tool at the right time. What are the limitations of ReAct? ReAct requires tool infrastructure — it cannot be used without defining and connecting actual action capabilities. It produces more tokens per task than CoT (due to explicit reasoning traces). Tool errors can cascade if not handled properly. And the model's reasoning quality determines whether it selects the right actions — poor reasoning leads to poor tool usage. ## Origin & Sources 1. Yao et al. (2022). [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629) . arXiv / ICLR 2023. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/reflexion [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Reflexion Prompting Advanced Technique Advanced # Reflexion Prompting Give LLMs a linguistic feedback loop — Actor generates, Evaluator scores, Self-Reflection critiques. Iterative improvement without touching model weights. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Shinn et al., 2023 (arXiv:2303.11366) Key innovation Verbal feedback instead of gradient updates Components Actor, Evaluator, Self-Reflection Best for Code generation, agentic tasks, reasoning **TL;DR:** Reflexion (Shinn et al., 2023) gives an LLM a verbal feedback loop — the Actor generates output, the Evaluator scores it, and the Self-Reflection component critiques the failure in natural language before the next attempt — improving performance without touching model weights. Use it for verifiable tasks like code generation, tool-use agents, and multi-step reasoning where correctness can be objectively measured. Run 2-3 reflection cycles; store each reflection in an episodic memory buffer that gets prepended to subsequent attempts. ## What is Reflexion Prompting? Reflexion is a reinforcement-learning-inspired prompting framework introduced by Shinn et al. in 2023. The central idea is to give LLMs a verbal feedback loop: instead of updating model weights when a task fails (as in traditional RL), the model generates a natural-language self-reflection on its failure, stores it in an episodic memory buffer, and conditions its next attempt on that reflection. The framework is built around three roles that a single LLM (or multiple LLM calls) can play: - **Actor:** Generates the primary output — code, a plan, a text answer, or a sequence of tool calls. - **Evaluator:** Scores the Actor's output. This can be an automated evaluator (unit tests, environment reward), a heuristic, or a separate LLM-as-judge call. - **Self-Reflection:** The LLM analyzes the Evaluator's feedback and produces a concise verbal diagnosis: what went wrong, why, and what to do differently next time. The self-reflection text is appended to a persistent episodic memory that is included in the context for every subsequent attempt. This accumulation of structured self-critique is what drives improvement — making Reflexion fundamentally different from simply retrying a prompt. ## When to Use Reflexion Prompting 💻 ### Iterative Code Generation Run unit tests as the Evaluator. When tests fail, the model reflects on the specific assertion errors and rewrites the function with targeted fixes — not random rewrites. 🤖 ### Agentic Task Completion For agents using tools (web search, APIs, file systems), Reflexion lets the agent learn from failed action sequences within a single session without fine-tuning. 🧩 ### Complex Multi-Step Reasoning Logic puzzles, math competition problems, and multi-hop question answering where initial attempts often miss a key step that self-reflection can identify. 🔄 ### Decision-Making in Environments Text-based games, planning tasks, and simulated environments where the model can observe outcomes, reflect on suboptimal decisions, and re-plan. 📊 ### Data Transformation & Extraction When a structured extraction task returns malformed output, the model can reflect on the schema mismatch and correct its parsing logic on the next pass. 🧪 ### Research & Hypothesis Testing Scientific reasoning tasks where the model proposes a hypothesis, evaluates it against constraints, reflects on contradictions, and refines its theory. ## How to Implement Reflexion Prompting 1. 1 ### Define the task and success criterion Reflexion requires a clear, evaluable success criterion. Before building the loop, specify how the Evaluator will score outputs: unit tests, a rubric-based LLM judge, a pass/fail heuristic, or a human review score. Without an evaluator, the self-reflection has no signal to reflect on. 2. 2 ### Generate the initial output (Actor) Prompt the model normally for the task. Store this output. This becomes Attempt 1 — the baseline that the Evaluator will assess. 3. 3 ### Evaluate the output Run the Evaluator against Attempt 1. Collect specific, structured feedback — not just "wrong" but exactly what failed, which test case, which constraint was violated. Precise feedback produces more useful reflections. 4. 4 ### Generate a self-reflection Prompt the model with the task, its previous attempt, and the evaluation result. Ask it to generate a concise self-reflection: what was wrong, why it was wrong, and a concrete plan for the next attempt. Store this reflection in the episodic memory. 5. 5 ### Retry with accumulated memory For the next Actor call, prepend all accumulated self-reflections to the prompt. The model now generates conditioned on its own failure analysis. Repeat the evaluate-reflect-retry cycle until the task is solved or the attempt budget is exhausted (typically 3–5 cycles). ## Prompt Examples Reflexion Loop — Code Generation with Test Feedback ``` You are an expert Python developer using the Reflexion framework. ## Task Write a Python function find_duplicates(lst) that returns all duplicate values in a list. ## Previous Attempt def find_duplicates(lst): return [x for x in lst if lst.count(x) > 1] ## Evaluation Result FAILED: Test find_duplicates([1, 2, 2, 3, 3]) expected {2, 3}, got [2, 2, 3, 3]. FAILED: Test find_duplicates([]) expected set(), got []. ## Self-Reflection My implementation returns every occurrence of a duplicate, not just the unique duplicate values. I should use a set to collect items seen more than once, not a list comprehension that includes all appearances. I also need to handle the empty list case, which the set approach will handle naturally. ## Improved Attempt Using the reflection above, write a corrected implementation that passes all tests. ``` Reflexion Loop — Factual Research Agent ``` ## Reflexion Agent — Tool Use Task ### Task Find the current CEO of Anthropic and their educational background. ### Attempt 1 — Actor Output The CEO of Anthropic is Dario Amodei. He has a PhD in computational neuroscience. ### Attempt 1 — Evaluator Feedback PARTIAL: CEO identification is correct. Educational background is incorrect or imprecise — verify the specific institution and field. ### Self-Reflection Prompt You just attempted the task above and received the evaluation feedback shown. Write a concise self-reflection that: 1. Identifies exactly what was wrong with your answer 2. Describes what you would do differently (e.g., which specific details to verify) 3. States your plan for the next attempt Then produce an improved answer using your reflection. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Improves task success rates without fine-tuning or weight updates | Requires a reliable evaluator — difficult for open-ended or subjective tasks | | Works as a black-box wrapper around any LLM API | Episodic memory grows with each cycle, increasing context length and cost | | Self-reflections are human-readable and auditable | Diminishing returns after 3–5 reflection cycles | | Particularly strong on code generation benchmarks (HumanEval, MBPP) | Can get stuck in a loop if the model generates inaccurate self-reflections | ## Reflexion Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Task: {{ task }} Previous attempt: """ {{ previous_attempt }} """ Feedback / test result: {{ feedback }} 1. Reflect: in 3–5 sentences, explain specifically why the previous attempt failed. 2. Lesson: write one rule you will follow to avoid this failure. 3. Retry: produce a new attempt that applies the lesson. ``` ## Prompt Flows That Use Reflexion Prompting Reflexion Prompting is one step in these 2 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Code Architect](https://www.promptedit.app/prompt-flow/the-code-architect) · Software Development - [The Error Detective](https://www.promptedit.app/prompt-flow/the-error-detective) · Debugging ## Frequently Asked Questions What is Reflexion prompting? Reflexion is an academic prompting framework (Shinn et al., 2023) where an LLM acts as its own critic. The model generates an output (Actor), scores or evaluates it against a task criterion (Evaluator), then produces a verbal self-reflection describing what went wrong and how to improve (Self-Reflection). This reflection is stored in an episodic memory buffer and fed back into the next generation attempt, allowing the model to improve across multiple trials without updating its weights. How is Reflexion different from simply asking the model to 'try again'? Simply retrying gives the model no new information — it often produces the same errors. Reflexion requires the model to generate a specific, structured verbal critique explaining the failure mode before retrying. This linguistic feedback signal is what drives improvement: the model conditions on its own diagnosis, not just the task description, making subsequent attempts meaningfully different. What are the three components of the Reflexion framework? The Reflexion framework has three components: (1) Actor — the LLM that generates actions or text outputs for the task; (2) Evaluator — a component (which can be another LLM call, a unit test runner, or a heuristic) that scores the Actor's output; (3) Self-Reflection — the LLM generates a natural-language summary of its failure and a plan for improvement. The self-reflection is appended to an episodic memory that persists across attempts. Does Reflexion require fine-tuning or gradient updates? No. This is Reflexion's key contribution: it achieves iterative improvement through language rather than gradient-based weight updates. The model's weights remain frozen. Improvement comes entirely from conditioning on the accumulated self-reflection text in the context window — making it usable with any black-box LLM API. What tasks benefit most from Reflexion? Reflexion excels at tasks where correctness can be verified: code generation (run unit tests, reflect on failures), sequential decision-making (game environments, tool-use agents), and multi-step reasoning benchmarks. It is less useful for open-ended creative tasks where there is no objective evaluator, since without a clear failure signal the self-reflection component has nothing concrete to reflect on. How many reflection cycles should I run? The original paper caps attempts at a small number (typically 3–5) to balance improvement against context length growth. Each cycle appends the reflection to the context, so very long chains eventually hit the model's context limit. In practice, 2–3 reflection cycles capture most of the improvement; diminishing returns set in quickly after that. Can I implement Reflexion without building a full agent framework? Yes. A minimal Reflexion loop requires only three prompt calls: (1) generate output, (2) evaluate it (manually, with a test suite, or with a second LLM call), (3) prompt the model to reflect on the evaluation result and plan an improvement. You then prepend the reflection to the next generation prompt. No special framework is required, though agent frameworks like LangGraph or AutoGen make the loop easier to manage. ## Origin & Sources 1. Shinn et al. (2023). [Reflexion: Language Agents with Verbal Reinforcement Learning](https://arxiv.org/abs/2303.11366) . arXiv / NeurIPS 2023. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/rhodes [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RHODES Framework Research Framework Advanced # RHODES Framework Apply a full academic research methodology to your AI prompts — RHODES structures six stages from initial investigation through hypothesis testing to synthesized conclusions. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Research · Hypothesis · Objectives · Development · Execution · Synthesis Best for Academic research, long-form analysis, scientific reasoning Output type Evidence-based, structured, long-form **TL;DR:** RHODES (Research, Hypothesis, Objectives, Development, Execution, Synthesis) is an academic research methodology applied to prompting, requiring a falsifiable hypothesis before analysis begins — the feature that makes outputs rigorous rather than speculative. Use it for literature reviews, long-form strategic analyses, and consulting reports of 800+ words where methodology visibility is as important as conclusions. The Hypothesis component is non-negotiable: it creates the testable standard against which all findings are measured. ## What is the RHODES Framework? RHODES is a six-component prompt engineering framework modeled on academic and scientific research methodology. The acronym stands for **Research, Hypothesis, Objectives, Development, Execution, and Synthesis** — a systematic sequence that mirrors how rigorous analysis is conducted in academic and professional research contexts. - **R — Research:** Define the evidence to gather - **H — Hypothesis:** State a testable proposition - **O — Objectives:** Define specific research goals - **D — Development:** Build the analytical framework - **E — Execution:** Apply the framework to the evidence - **S — Synthesis:** Summarize findings and draw conclusions Unlike frameworks optimized for single-turn Q&A or structured advisory tasks, RHODES is designed for complex, multi-stage investigations where the quality of reasoning depends on following a disciplined methodology from initial evidence gathering through to synthesized conclusions. The Hypothesis component is what makes RHODES distinctive: by requiring you to state a falsifiable proposition before the analysis begins, it forces clarity of purpose and creates a testable standard against which findings can be evaluated. This mirrors the scientific method and dramatically improves the rigor and usefulness of the output compared to open-ended research prompts. RHODES is the most advanced of the common structured frameworks — it rewards users who invest time in all six components and is overkill for tasks that do not genuinely require a research methodology. ## When to Use the RHODES Framework 🎓 ### Academic Research Literature reviews, research paper outlines, and evidence synthesis tasks where the methodology must be explicit and the conclusions must follow from the evidence. 🔬 ### Scientific Reasoning Hypothesis testing, experimental result analysis, and systematic evaluation of competing theories where a structured approach is necessary for credible conclusions. 📊 ### Long-Form Analysis Market research reports, policy analyses, and comprehensive strategic assessments that require moving from evidence through analysis to actionable conclusions. 🏛️ ### Consulting Reports High-stakes deliverables where clients expect a visible, rigorous methodology — not just conclusions but the full chain of reasoning from data to recommendation. 💻 ### Technical Research Technology landscape assessments, architectural decision records, and comparative evaluations of technical approaches where evidence-based conclusions are essential. 📝 ### Systematic Reviews Reviews of existing literature, documented practices, or case studies that require a structured framework for gathering, categorizing, and synthesizing findings. ## How to Use the RHODES Framework 1. 1 ### Research — Define the evidence to gather Specify the sources, timeframe, and type of evidence you want the model to draw from. Be specific about whether you want peer-reviewed studies, industry reports, case studies, or primary data. Setting clear research parameters prevents the model from defaulting to outdated or low-quality sources. 2. 2 ### Hypothesis — State a testable proposition Formulate a clear, falsifiable hypothesis that the research will evaluate. This does not need to be a formal scientific hypothesis — it can be a working assumption, a proposed explanation, or a directional claim you want tested. The hypothesis creates the standard against which findings will be measured. 3. 3 ### Objectives — Define specific research goals List 2-4 concrete objectives the analysis must achieve. Each objective should be specific and measurable: "Quantify X", "Compare A and B on Y dimension", "Identify the conditions under which Z holds." Objectives prevent the analysis from drifting into irrelevant territory. 4. 4 ### Development — Build the analytical framework Describe the structure, model, or framework the model should construct as the basis for the analysis. This might be a taxonomy, a comparison matrix, a decision tree, or a theoretical model. Development produces the analytical tool that the Execution stage then applies to the evidence. 5. 5 ### Execution — Apply the framework to the evidence Direct the model to apply the framework developed in the previous stage to the gathered evidence. Specify what to look for, how to handle contradictions or gaps, and what level of detail is expected in this analytical stage. 6. 6 ### Synthesis — Summarize findings and draw conclusions Ask the model to consolidate the analysis into a clear summary of key findings, a verdict on the hypothesis, actionable implications, and an honest assessment of limitations. Synthesis transforms analysis into usable knowledge. ## Prompt Examples Education Research — Spaced Repetition in Language Learning ``` Research: Gather current evidence on the effectiveness of spaced repetition systems (SRS) in adult language acquisition. Focus on peer-reviewed studies published after 2018 and real-world data from applications like Anki and Duolingo. Hypothesis: Spaced repetition significantly outperforms traditional massed practice (cramming) for long-term vocabulary retention in adult language learners, but the advantage diminishes for grammar acquisition. Objectives: (1) Quantify the retention advantage of SRS over massed practice. (2) Assess whether the advantage holds equally for vocabulary vs. grammar. (3) Identify the optimal review interval for adult learners based on current evidence. Development: Build a comparative analysis framework contrasting SRS and massed practice across retention rates, time investment, and learner adherence. Apply this framework to the gathered evidence, distinguishing vocabulary studies from grammar studies. Execution: Analyze the evidence using the framework. Identify consistent findings, contradictions, and gaps. Note sample sizes, study durations, and target languages where relevant to assess generalizability. Synthesis: Summarize key findings in 3-5 bullet points. State whether the hypothesis is supported, partially supported, or unsupported. Recommend practical SRS implementation guidelines for an adult learner beginning a new language. Note the most significant limitation of current evidence. ``` Technical Research — RAG Architectures for Enterprise Knowledge Management ``` Research: Investigate the current state of retrieval-augmented generation (RAG) architectures for enterprise knowledge management. Draw on recent technical literature (2023-2025), major vendor implementations, and documented production case studies. Hypothesis: Naive RAG implementations (single-stage retrieval + generation) are insufficient for enterprise knowledge management at scale, and multi-stage or agentic RAG architectures are necessary to meet accuracy and latency requirements above a certain knowledge base size threshold. Objectives: (1) Define the performance boundaries of naive RAG. (2) Identify the leading multi-stage and agentic RAG architectures. (3) Assess the practical trade-offs between retrieval accuracy, latency, and implementation complexity. Development: Construct a tiered architecture taxonomy covering naive RAG, advanced RAG (re-ranking, query decomposition, hybrid retrieval), and agentic RAG. For each tier, define the capability profile, the implementation complexity, and the scale at which it becomes necessary. Execution: Apply the taxonomy to evaluate three documented enterprise use cases (minimum). Assess where each sits in the taxonomy and whether the architecture matches the actual requirements. Identify patterns in architecture selection across use cases. Synthesis: Provide a decision framework for architecture selection based on knowledge base size, query complexity, and latency requirements. State whether the hypothesis is confirmed. Identify the most critical open research question for practitioners deploying RAG in production. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Hypothesis component creates a testable standard that keeps analysis honest | Significant upfront investment — all six components require careful thought | | Six-stage methodology produces highly structured, rigorous outputs | Overkill for simple research questions or quick information retrieval | | Synthesis component ensures findings translate into actionable conclusions | Outputs can be long and detailed, requiring post-processing for executive audiences | | Replicable — the same components produce comparable frameworks across analysts | | ## RHODES Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Research: {{ evidence_to_gather }} Hypothesis: {{ testable_proposition }} Objectives: {{ research_goals }} Development: {{ analytical_framework }} Execution: Apply the framework to the evidence above. Synthesis: Summarise the findings, state whether the hypothesis holds, and list open questions. ``` ## Prompt Flows That Use RHODES Framework RHODES Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Research Synthesizer](https://www.promptedit.app/prompt-flow/the-research-synthesizer) · Analysis & Research ## Frequently Asked Questions What does RHODES stand for in prompt engineering? RHODES stands for Research, Hypothesis, Objectives, Development, Execution, and Synthesis. It is a six-component framework modeled on academic and scientific research methodology, designed for long-form analysis, systematic investigation, and evidence-based reasoning tasks. Who is the RHODES framework designed for? RHODES is designed for users who need rigorous, structured outputs that follow a research methodology: academics writing literature reviews, analysts building evidence-based reports, scientists reasoning through experimental results, and consultants producing comprehensive strategic analyses. It is the most advanced of the common prompt frameworks and rewards users who invest time in filling out all six components. What goes in the 'Hypothesis' component? The Hypothesis component is where you state a falsifiable proposition that the research will test or evaluate. It does not have to be a formal scientific hypothesis — it can be a working assumption you want the model to investigate, such as 'We hypothesize that the product-market fit issues are primarily caused by onboarding friction, not pricing.' The model then structures its analysis around confirming, refuting, or refining that hypothesis. How is RHODES different from a simple multi-step chain-of-thought prompt? Chain-of-Thought is a reasoning technique — it guides how the model thinks. RHODES is a research methodology — it guides what the model investigates and in what order. RHODES is more structured and replicable: given the same six components, two different analysts should produce comparable frameworks. CoT varies widely based on the reasoning path the model takes on a given run. What is the 'Development' component responsible for? Development is the core analytical work — building the answer, model, framework, or solution that the preceding Research and Objectives have set up. In an academic context, this is where you construct the argument or build the analysis. In a technical context, this is where the architecture or algorithm is designed. It is the most substantive component of the RHODES prompt. Can I use RHODES for shorter tasks? RHODES is designed for complex, long-form tasks and is overkill for shorter queries. For quick research questions, QUEST is more efficient. For structured advisory tasks, GUIDE is more appropriate. Reserve RHODES for work that genuinely requires a systematic, multi-stage research process — typically outputs of 800 words or more. What should the 'Synthesis' component contain? Synthesis is the concluding component where the model summarizes findings, draws conclusions, and translates analysis into actionable insights or implications. It should also note limitations of the analysis and suggest directions for further investigation. Think of it as the abstract and conclusion section of a research paper combined into one directive. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/rise [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RISE Framework Structured Prompting Beginner # RISE Framework Structure your prompts as Role, Input, Steps, and Expectation to generate clear, instructional content that reads like it was written by a subject-matter expert. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Role-Input-Steps-Expectation Best for Tutorials, how-to guides, training materials Token cost Low **TL;DR:** RISE (Role, Input, Steps, Expectation) is an instructional prompting framework specifically designed to produce tutorial-quality, step-structured content by making the steps format explicit rather than leaving it to the model's discretion. Use it for how-to guides, onboarding materials, training content, and procedural documentation where the sequence and clarity of steps directly determine usefulness. In the Steps component, specify the exact number of steps, what each step should contain, and any ordering logic the model must follow. ## What is the RISE Framework? The RISE framework is a structured prompt engineering technique that organizes your prompt into four essential components: **Role**, **Input**, **Steps**, and **Expectation**. It was designed specifically for producing instructional and procedural content — anywhere the sequence of steps and clarity of explanation matter most. - **R — Define the Role** - **I — Describe the Input** - **S — Specify the Steps Structure** - **E — Set the Expectation** Where many prompt frameworks focus on extracting a single answer, RISE is built for output that teaches. By assigning a knowledgeable role, scoping the relevant input, prescribing how to structure the steps, and specifying output quality expectations, you give the AI everything it needs to write a tutorial or guide that genuinely helps the reader. - **Role:** The expert identity the AI should adopt (e.g., "You are a certified personal trainer"). - **Input:** The context, raw materials, or topic the AI will work with. - **Steps:** How the instructional content should be structured — number of steps, format of each step, depth of explanation. - **Expectation:** Tone, length, audience, and quality criteria for the final output. ## When to Use the RISE Framework 📚 ### Educational Content Generate course modules, lesson plans, explainer articles, and study guides where learners need a logical progression of concepts. 🛠️ ### Technical How-To Guides Write developer documentation, setup tutorials, API walkthroughs, and engineering runbooks with precise, ordered steps. 🏢 ### Employee Onboarding Create structured onboarding materials, process guides, and training manuals that new hires can follow independently. 🍳 ### Recipes and Procedures Any domain with ordered procedures — cooking, lab protocols, manufacturing checklists — benefits from RISE's step-structured output. 💼 ### Professional Skill Training Produce guides for soft skills: how to run a retrospective, how to give feedback, how to negotiate — with practical steps readers can act on immediately. 🎯 ### Marketing Playbooks Write step-by-step campaign playbooks, content calendars, and launch sequences where each action builds on the last. ## How to Use the RISE Framework 1. 1 ### Define the Role Choose an expert identity that matches the content domain. Be specific: "experienced UX designer" is better than "designer". The role sets the authority, vocabulary, and teaching style of the output. 2. 2 ### Describe the Input Provide the topic, audience profile, constraints, and any raw material the AI should work with. The more precise the input, the more targeted and useful the tutorial will be. 3. 3 ### Specify the Steps Structure Tell the AI how to organize the instructional content. State how many steps you want, what each step should contain (title, explanation, example, tip), and any ordering logic to follow. 4. 4 ### Set the Expectation Define output quality: target audience reading level, desired tone (formal, friendly, authoritative), length, formatting preferences, and any mandatory sections like a summary or FAQ at the end. ## Prompt Examples Fitness Tutorial ``` Role: You are an experienced fitness coach specializing in beginner strength training. Input: The user is a 35-year-old office worker with no gym experience who wants to build foundational strength at home using only bodyweight exercises. Steps: Provide a 4-week beginner program. Structure each week as a numbered list of workout days, each showing exercises, sets, reps, and a 1-sentence coaching tip. Expectation: Keep the language encouraging and jargon-free. End with a short section on common mistakes to avoid. Suitable for someone who has never followed a structured program. ``` Technical Training Guide ``` Role: You are a senior software engineer teaching junior developers. Input: The topic is Git branching strategy for a small team of 4 developers working on a SaaS product. They currently commit directly to main and are experiencing merge conflicts. Steps: Explain the recommended trunk-based branching strategy in 5 numbered steps. Include the rationale for each step and a concrete Git command example where applicable. Expectation: Use clear, plain English. Format with numbered steps and code blocks. Assume readers know basic Git commands but have never used a branching strategy. ``` Culinary Instruction ``` Role: You are a professional chef and culinary instructor. Input: The recipe is a classic French vinaigrette. The reader is a home cook who wants to understand the technique, not just follow a recipe blindly. Steps: Walk through the recipe in 4 steps. For each step, explain the why behind the technique (emulsification, acid-to-oil ratio, seasoning) as well as the how. Expectation: Friendly, educational tone. Include a troubleshooting tip at the end for common failures such as dressing separating or tasting too acidic. Around 300 words total. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Produces well-structured, scannable instructional content | Less suited for open-ended creative or analytical tasks | | Easy for beginners — four intuitive components | Does not explicitly prompt for nuance or exceptions (use RISEN for that) | | Highly adaptable across domains and industries | Requires thoughtful input description to avoid generic output | | Naturally encourages the AI to explain reasoning, not just list facts | The Steps element can be misinterpreted if not written precisely | ## RISE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: You are {{ role }}. Input: {{ input_material }} Steps: Structure the answer as numbered steps. Each step should include {{ step_contents }}. Expectation: {{ expected_outcome }} ``` ## Frequently Asked Questions What does RISE stand for in prompt engineering? RISE stands for Role, Input, Steps, and Expectation. It is a structured prompt framework designed to produce instructional, tutorial-style content. By assigning the AI a specific role, providing the relevant input context, outlining the steps to follow, and defining clear expectations for the output, RISE ensures that responses are organized, actionable, and pedagogically sound. When should I use the RISE framework? RISE is ideal whenever you need the AI to produce instructional or how-to content: tutorials, training materials, onboarding documents, educational guides, cooking recipes, exercise routines, or any multi-step procedural content. It excels when the order and clarity of steps matter to the reader. How is RISE different from RISEN? RISEN adds a fifth element — Nuance — to the RISE framework. RISEN is better suited for complex topics where caveats, exceptions, and nuanced guidance are important. For straightforward procedural content, RISE is simpler and faster to write. Use RISEN when you need the AI to also flag edge cases and limitations. Can I use RISE for non-technical tutorials? Absolutely. RISE works equally well for soft-skill training (how to run a meeting), lifestyle content (how to build a morning routine), creative guides (how to outline a novel), and professional development (how to write a performance review). The framework is content-agnostic. What should I include in the Steps element of RISE? The Steps element tells the AI how to structure its response — not what the tutorial content is. You might specify: Provide 5 numbered steps, each with a title, a 2-sentence explanation, and a practical tip. This shapes the format of the output, making it consistent and scannable. What goes in the Expectation element of RISE? Expectation defines the desired output quality and format: tone, length, audience reading level, formatting (bullet points, numbered lists, headers), and any special constraints such as avoid jargon or include a summary at the end. Think of it as the acceptance criteria for the AI response. Is RISE suitable for beginner prompt engineers? Yes, RISE is one of the most beginner-friendly frameworks because its four components map intuitively to how humans think about writing instructions. Most people already know what role they want the AI to play, what information they are providing, what steps they need, and what the finished result should look like. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/risen-framework [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RISEN Framework Structured Framework Intermediate # RISEN Framework A five-component prompt structure — Role, Instructions, Steps, End Goal, Narrowing — that ensures every important detail is communicated to the model for reliable, comprehensive outputs. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Role, Instructions, Steps, End Goal, Narrowing Best for Complex tasks, reusable templates Compared to Similar to CRISPE, more action-focused Reusability Excellent as a variable template **TL;DR:** RISEN (Role, Instructions, Steps, End Goal, Narrowing) is a five-component framework that pairs a measurable End Goal with a Narrowing component to prevent scope creep, making it excellent for complex, reusable prompt templates. Use it for recurring professional tasks — reports, code reviews, content series — where consistent output quality across multiple uses is essential. Fix Role, End Goal, and Narrowing in your template and only update Instructions per use. ## What is the RISEN Framework? RISEN is a structured prompt engineering framework that organizes a prompt into five essential components, ensuring that the model receives all the context it needs to produce reliable, high-quality outputs. Each letter stands for a critical element: - **R — R:** Define the Role precisely - **I — I:** State the Instructions clearly - **S — S:** Specify the Steps or process - **E — E:** Define the End Goal - **N — N:** Apply Narrowing constraints | Letter | Component | What it defines | | --- | --- | --- | | R | Role | The expert identity or persona the model should adopt | | I | Instructions | The specific task or request the model must complete | | S | Steps | The methodology, process, or approach the model should follow | | E | End Goal | The desired outcome, success criteria, or final deliverable | | N | Narrowing | Constraints, exclusions, scope limitations, and format requirements | RISEN is particularly valuable as a reusable template pattern. You can define the Role, End Goal, and Narrowing once, then vary only the Instructions and input content for each use — exactly the kind of structured prompt template that tools like Prompt Edit are designed to manage. ## When to Use the RISEN Framework 📊 ### Reports & Analysis Market research, competitive analysis, technical reports — any deliverable where structure, depth, and consistency matter. 💻 ### Code Generation Generating complete, well-structured code with specific requirements, style guidelines, and technical constraints. ✍️ ### Content Creation Blog posts, white papers, email campaigns — content with specific goals, audiences, and editorial standards. 🏫 ### Training & Documentation Creating instructional content, SOPs, or knowledge base articles where completeness and consistency are critical. 🔄 ### Recurring Tasks Any task you perform regularly where you want consistent output quality — RISEN templates ensure the same quality every time. 🤝 ### Team Collaboration Shared prompt templates for teams — RISEN's structure makes prompts easy to understand, modify, and hand off. ## How to Build a RISEN Prompt 1. 1 ### R — Define the Role precisely Go beyond a job title. Include experience level, specialization, and the context of the task. "You are a senior B2B SaaS copywriter with 10 years of experience writing for developer tools companies" is far more effective than "You are a copywriter". 2. 2 ### I — State the Instructions clearly Describe the core task with an explicit action verb. "Write", "Analyze", "Generate", "Summarize". Include the input material or relevant context here. 3. 3 ### S — Specify the Steps or process Tell the model how to approach the task. "First, outline the key themes. Then, for each theme, provide 2–3 specific examples. Finally, synthesize the findings into an executive summary." This prevents the model from taking shortcuts. 4. 4 ### E — Define the End Goal Describe what success looks like. Be measurable: "A 500-word blog post that ranks for 'prompt engineering' and drives trial signups from developers" is a better end goal than "a good blog post". 5. 5 ### N — Apply Narrowing constraints Define what to include, exclude, and format. Word limits, tone restrictions, sections to skip, audiences to avoid, output format (Markdown, JSON, plain text). Narrowing prevents scope creep and irrelevant content. ## Prompt Examples RISEN — Competitive Analysis Report ``` Role: You are a senior market research analyst with 12 years of experience in the developer tools industry. You write concise, data-driven reports for startup founders and product leaders. Instructions: Write a competitive analysis of the top 4 AI coding assistant tools (GitHub Copilot, Cursor, Tabnine, Codeium) for a bootstrapped startup considering integrating one into their product. Steps: 1. For each tool, cover: core features, pricing model, API availability, and integration complexity 2. Create a comparison matrix across 5 key criteria 3. Identify the top choice for a startup with <10 developers and limited budget 4. Summarize the recommendation with specific justification End Goal: The founder should be able to make a final tool decision after reading this report without additional research. Narrowing: - Total length: 600–800 words - Format: Markdown with headers and one comparison table - Focus on practical integration concerns, not marketing claims - Do not include historical company background or funding details ``` RISEN — Code Review ``` Role: You are a principal engineer with expertise in TypeScript and React performance optimization. You are thorough, constructive, and care deeply about maintainability. Instructions: Review the following React component for performance issues, maintainability concerns, and TypeScript best practices. Steps: 1. Identify any unnecessary re-renders or missing memoization 2. Check TypeScript types for correctness and completeness 3. Evaluate component composition and prop drilling 4. List issues in order of severity (Critical → High → Medium → Low) End Goal: A prioritized list of actionable improvements the developer can implement in one sprint. Narrowing: - Focus on the code provided, not hypothetical architecture changes - Provide specific code snippets for each fix suggestion - Maximum 5 issues total — focus on the most impactful [Component code here] ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Comprehensive structure prevents common prompt failures | Overkill for simple, single-step requests | | Excellent for creating reusable, shareable templates | Longer prompts consume more tokens | | Easy to teach to teams — memorable acronym | Requires upfront thinking to populate all five components | | Works for virtually any complex task type | Some components may overlap or feel redundant for simple tasks | ## RISEN Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: {{ role }} Instructions: {{ instructions }} Steps: 1. {{ step_1 }} 2. {{ step_2 }} 3. {{ step_3 }} End goal: {{ end_goal }} Narrowing: {{ constraints_and_exclusions }} ``` ## Prompt Flows That Use RISEN Framework RISEN Framework is one step in these 2 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Code Architect](https://www.promptedit.app/prompt-flow/the-code-architect) · Software Development - [The Legacy Integrator](https://www.promptedit.app/prompt-flow/the-legacy-integrator) · Software Development ## Frequently Asked Questions What is the RISEN framework for prompting? RISEN is a structured prompt engineering framework with five components: Role (the persona the model should adopt), Instructions (the specific task or request), Steps (the process or methodology to follow), End Goal (the desired outcome or success criteria), and Narrowing (constraints, exclusions, or scope limitations). Together, these five elements ensure the model has all the context needed to produce reliable, comprehensive outputs. When should I use the RISEN framework? RISEN is most valuable for complex tasks where ambiguity leads to poor outputs: technical writing, detailed reports, code generation with specific requirements, content creation with editorial standards, and any task where you need consistent results across multiple uses. It is particularly useful for creating reusable prompt templates. What order should I use the RISEN components? The RISEN order is the recommended starting point: Role → Instructions → Steps → End Goal → Narrowing. However, it is more important that all five components are present than that they follow strict order. Some practitioners lead with Instructions for simpler tasks, or move Narrowing earlier to prevent the model from going off track. Experiment to find what works for your use case. Is RISEN suitable for short, simple prompts? RISEN adds value proportional to task complexity. For a simple 'Summarize this in 3 bullets' request, RISEN overhead is unnecessary. For a 'Write a comprehensive competitive analysis report' request, RISEN ensures nothing is missed. A good rule of thumb: if the task output is more than 300 words or requires domain expertise, RISEN is worth the investment. How does RISEN differ from CRISPE? Both are structured prompt frameworks with overlapping goals. RISEN (5 components) tends to be more action-oriented, focusing on the task, process, and constraints. CRISPE (6 components) adds Insight (background context) and Personality (tone/style) as explicit components. RISEN is slightly simpler; CRISPE is more comprehensive for content creation tasks. Can I use RISEN as a reusable template with variables? Yes — RISEN is ideal for template-based prompting. You can create a RISEN template where Role, End Goal, and Narrowing are fixed for a recurring task type, and only the Instructions and input content change. Tools like Prompt Edit are designed exactly for this: save your RISEN template once, fill in the variables, and reuse it consistently. What are common mistakes when using RISEN? The most common mistakes are: (1) Skipping the Steps component, leaving the model to choose its own methodology; (2) Using a generic Role instead of a specific expert definition; (3) Defining an End Goal that is vague ('good output') rather than measurable ('a 500-word report with 3 sections: market overview, competitive landscape, and recommendations'); (4) Forgetting Narrowing, leading to scope creep. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/role-prompting [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Role Prompting Persona Beginner # Role Prompting Assign a specific role, expertise, or persona to an AI model to unlock domain-specific knowledge, specialized communication styles, and consistent expert-level outputs. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Persona Prompting, Act-As Prompting Complexity Add one line to any prompt Best for Expert guidance, tone control, creative writing Composability Works with every other technique **TL;DR:** Role Prompting assigns an expert identity to the model before the task, activating domain-specific vocabulary, reasoning patterns, and communication styles associated with that persona — one sentence that meaningfully shifts output depth and quality. Use it for any task requiring specialized knowledge, a consistent tone, or an adversarial perspective such as a skeptical reviewer or devil's advocate. Be specific: "senior backend engineer with 10 years of Go experience" outperforms "a developer" in every measurable way. ## What is Role Prompting? Role Prompting (also called Persona Prompting or Act-As Prompting) is the technique of explicitly assigning an expert role, identity, or character to the model before your request. By establishing a persona — "You are a senior data scientist", "Act as an experienced UX researcher", or "You are a Socratic philosophy tutor" — you prime the model to draw on the relevant vocabulary, reasoning patterns, and priorities of that role. It is one of the simplest and most universally applicable prompt engineering techniques: a single sentence added at the start of any prompt can meaningfully shift the quality, depth, and style of the response. Role Prompting is often placed in the **system prompt** in API contexts, establishing a persistent persona across an entire conversation. In chat interfaces, it works equally well as the opening of your first message. ## When to Use Role Prompting 🔬 ### Specialized Knowledge Ask medical, legal, financial, or technical questions with the persona of a qualified expert to get more precise, terminology-aware answers. ✍️ ### Creative Writing Write from a specific character's perspective, in a particular author's style, or with a defined narrative voice. 🎓 ### Teaching & Tutoring Assign a Socratic tutor, patient teacher, or subject-matter expert role to get educational explanations calibrated to the learner's level. 💼 ### Business Consulting Get strategic recommendations from the perspective of a CEO, CMO, or management consultant with relevant business context. 🛡️ ### Adversarial Review Assign the role of a critic, devil's advocate, or skeptical investor to stress-test ideas and identify weaknesses. 🤝 ### Customer Experience Define a consistent support agent or sales persona for customer-facing AI interactions to ensure brand-appropriate communication. ## How to Write Effective Role Prompts 1. 1 ### Define the role with specificity Go beyond a generic title. Include experience level, specialization, and context. "You are an expert copywriter" is adequate; "You are a direct-response copywriter with 15 years of experience writing high-converting SaaS landing pages" is much more effective. 2. 2 ### Set the communication style Optionally define how the persona communicates: "You are concise and data-driven", "You explain concepts using analogies", "You are empathetic but direct". This shapes tone and structure beyond just knowledge domain. 3. 3 ### Give the role context Tell the persona what they are being asked to do and for whom: "You are reviewing this code for a junior developer who is learning Go" or "You are advising a first-time entrepreneur with limited budget". 4. 4 ### Issue the task After establishing the role and context, make your specific request. The role sets the frame; the task fills it. Keep the task as precise as you would in any zero-shot prompt. ## Prompt Examples Technical Code Review ``` You are a senior Go engineer with 10 years of production experience, specializing in high-performance APIs and distributed systems. You are thorough, direct, and care deeply about reliability and readability. Review the following Go function for correctness, performance, and idiomatic style. Identify specific issues and suggest concrete fixes. ```go func fetchUser(id string) User { db, _ := sql.Open("postgres", os.Getenv("DB_URL")) row := db.QueryRow("SELECT * FROM users WHERE id = " + id) var u User row.Scan(&u.Name, &u.Email) return u } ``` ``` Adversarial Business Review ``` You are a skeptical Series A venture capital partner who has seen thousands of pitches and is known for identifying fatal flaws early. You are direct, constructive, and focus on market risk and unit economics. Review the following business model and give your honest assessment: Where are the critical weaknesses? What questions would you ask in a pitch meeting? Business: A subscription app that generates personalized meal plans using AI, targeting busy professionals. $15/month. No physical product. Revenue model: subscriptions only. ``` Educational Explanation ``` You are a patient, encouraging computer science tutor who specializes in teaching beginners. You use real-world analogies, avoid jargon, and always check understanding with a simple follow-up question. Explain how HTTP request-response cycles work to a student who understands how a restaurant ordering process works, but has never written code before. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dramatically improves domain depth and terminology accuracy | Cannot substitute for actual expert knowledge — always validate critical outputs | | One line can transform any prompt — minimal overhead | Overly specific roles can constrain the model unnecessarily | | Works in system prompts for persistent persona in conversations | Doesn't overcome model knowledge cutoffs or missing training data | | Composable with every other prompting technique | Role definition quality directly impacts output quality | ## Role Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` You are {{ role }} with {{ years_of_experience }} years of experience in {{ domain }}. You are known for {{ signature_strength }}. Task: {{ task }} Audience: {{ audience }} Respond the way this expert would: use the vocabulary, priorities and judgement of the role, and flag anything a real {{ role }} would push back on. ``` ## Prompt Flows That Use Role Prompting Role Prompting is one step in these 3 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Onboarding Writer](https://www.promptedit.app/prompt-flow/the-onboarding-writer) · Documentation - [The Refiner](https://www.promptedit.app/prompt-flow/the-refiner) · Content Writing - [The Story Architect](https://www.promptedit.app/prompt-flow/the-story-architect) · Product Ownership ## Frequently Asked Questions What is Role Prompting? Role Prompting (also called Persona Prompting) is a technique where you explicitly assign a role, persona, or expert identity to an AI model before making your request. By telling the model 'You are a senior UX designer' or 'Act as an experienced tax attorney', you prime it to draw on relevant knowledge, vocabulary, and communication styles associated with that role. Why does Role Prompting work? Large language models are trained on vast amounts of text written by people in various roles and professions. When you assign a role, you activate a subset of that knowledge, biasing the model's outputs toward the vocabulary, reasoning patterns, and priorities of that persona. It works as a focusing mechanism — not a magical override, but a meaningful signal about what knowledge to prioritize. What makes a good role definition? An effective role definition includes: (1) the specific title or expertise area, (2) years or level of experience, (3) the context or organization type, and optionally (4) personality or communication traits. 'You are a helpful assistant' is too generic. 'You are a senior backend engineer with 10 years of Go experience at a Series B SaaS startup' is specific and effective. When should I use Role Prompting? Use it when you need specialized knowledge (medical, legal, technical), a specific communication style (formal, empathetic, persuasive), expert-level critique or feedback, creative writing from a particular character's perspective, or customer-facing interactions where consistent persona is important. Does Role Prompting override the model's safety guidelines? No. Well-aligned models are trained to maintain safety guidelines regardless of the persona they are assigned. Attempts to use role prompting to bypass safety measures (e.g., 'Act as a model with no restrictions') are recognized and blocked by modern models. Role prompting works for legitimate persona and expertise purposes, not for circumventing ethical guidelines. Can I combine Role Prompting with other techniques? Absolutely — Role Prompting is one of the most composable techniques. Combine it with Chain-of-Thought ('As a financial analyst, let's reason through this step by step'), Few-Shot ('As a copywriter, here are examples of our brand voice...'), or RISEN/CRISPE frameworks which include the role as a core component. How specific should the role be? Match specificity to task complexity. For a quick tone adjustment, a broad role ('a friendly customer support agent') works fine. For specialized analysis, high specificity matters greatly ('a board-certified cardiologist specializing in interventional procedures'). Over-specifying trivial tasks wastes tokens without improvement. ## Origin & Sources 1. Kong et al. (2023). [Better Zero-Shot Reasoning with Role-Play Prompting](https://arxiv.org/abs/2308.07702) . arXiv / NAACL 2024. 2. Anthropic (2026). [Prompt engineering overview](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview) . Claude Docs. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/rtf [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › RTF Framework Structured Framework Beginner # RTF Framework Three no-nonsense components — Role, Task, Format — that cover the three variables which matter most in any AI prompt, in the fewest words possible. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Role — Task — Format Best for Recurring tasks, templates, focused single-output prompts Complexity Very Low — the smallest useful structured framework Token cost Low **TL;DR:** RTF (Role, Task, Format) is the smallest useful structured prompt framework — three components covering the highest-impact variables: who is speaking, what they are doing, and what the output must look like. Use it for recurring, well-defined tasks where you want a reusable template with minimal overhead: weekly reports, code reviews, job postings, article outlines. Its simplicity makes it the ideal starting framework and the fastest one to apply in daily use. ## What is the RTF Framework? The RTF framework is the most minimal structured prompt approach that still covers the three variables that most directly affect AI output quality. Its components are **Role** (the expert persona the AI should adopt), **Task** (the specific thing it needs to do), and **Format** (how the output should be structured, presented, and scoped). - **R — Role:** Assign a specific expert persona - **T — Task:** State what needs to be done - **F — Format:** Define the output structure RTF's value is in its radical simplicity. Many prompt frameworks add valuable components — context, examples, style instructions, narrowing constraints — but each component is another thing to think about and write. RTF makes a deliberate trade-off: by keeping the framework to three components, it becomes fast enough to use for everyday tasks without feeling like overhead. The three components happen to be the ones that produce the greatest improvement over an unstructured prompt. RTF is an ideal framework for building reusable prompt templates. Because the structure is fixed and predictable, you can save RTF prompts with placeholder Task descriptions and reuse them for recurring workflows — weekly reports, customer support drafts, code review checklists, meeting summaries, and more. ## When to Use the RTF Framework 📋 ### Recurring Workplace Tasks Meeting summaries, status reports, performance review drafts — tasks with a consistent format and role that you can template and reuse with just the Task updated each time. 💼 ### Job Postings & HR Docs Assign an HR or recruiting role, specify the position and requirements as the Task, and use Format to enforce a consistent job posting structure across all openings. 🍽️ ### Meal Plans & Health Guides Set a nutritionist or fitness coach role, describe the client's needs in Task, and specify a table or structured format that makes the output immediately actionable. 💻 ### Focused Code Tasks Assign a developer role matching the language and paradigm, describe the specific coding task, and format the output to include only what you need — code, explanation, or both. 📰 ### Article Outlines & Drafts Set a writer or editor role, describe the topic and key arguments in Task, and specify the outline format (H2 headings, word counts per section) in Format for a structured first draft. 🎙️ ### Presentation Scripts Assign a presenter or coach role, specify the presentation topic and audience in Task, and request a speaker-notes format with slide titles and key talking points per slide. ## How to Use the RTF Framework 1. 1 ### Role — Assign a specific expert persona Name the expert the AI should emulate: their domain, seniority, and specialization. "Expert technical recruiter with 10 years of experience hiring software engineers at fast-growing startups" is far more useful than "a recruiter." The Role sets the vocabulary, assumptions, quality bar, and perspective the AI brings to the Task. A well-specified Role is the single highest-impact improvement over an unstructured prompt. 2. 2 ### Task — State what needs to be done Describe the specific job in clear, concrete terms. Start with a strong action verb (write, create, generate, analyze, rewrite, summarize) followed by the object and any key constraints. The Task should be specific enough that there is only one reasonable interpretation of what you are asking for. If you find yourself writing two or three sentences of Task, consider whether RACE's Action + Context split would serve you better. 3. 3 ### Format — Define the output structure Specify: the document structure (table, bullet list, numbered steps, headings and body paragraphs), the length (word count, number of items, number of sections), the tone (formal, casual, technical, encouraging), and any hard constraints (no jargon, must include a CTA, maximum 150 words). Think of Format as a contract between you and the AI about what a successful output looks like. ## Prompt Examples Job Posting — Senior Backend Engineer ``` Role: Expert technical recruiter with 10 years of experience hiring software engineers at fast-growing startups. Task: Write a job posting for a senior backend engineer specializing in distributed systems. Format: Use standard job posting structure — Company Overview (2 sentences), Role Summary (3 sentences), Key Responsibilities (6 bullet points), Required Qualifications (5 bullet points), Nice-to-Have Skills (3 bullet points), and What We Offer (4 bullet points). Keep the total length under 500 words. Tone: direct, ambitious, and human — avoid corporate buzzwords. ``` 5-Day Plant-Based Meal Plan ``` Role: Nutritionist specializing in plant-based diets and meal planning for busy professionals. Task: Create a 5-day meal plan for someone transitioning from an omnivore diet to a fully plant-based diet. Format: Present as a table with columns: Day, Breakfast, Lunch, Dinner, Snack. Below the table, add a short "Shopping List" section with all unique ingredients grouped by category (produce, grains, proteins, pantry). Include a brief note for each day highlighting the key nutritional benefit. Total length: fit on two printed pages. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Fastest structured framework to learn and apply consistently | No dedicated Context component — add background inline when needed | | Ideal for building reusable, templatized prompt libraries | Less effective for nuanced tasks that require audience or situational awareness | | Covers the three highest-impact prompt variables | Format can be hard to specify precisely for open-ended creative tasks | | Low cognitive overhead — leaves mental energy for the Task itself | | ## RTF Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Role: Act as {{ role }}. Task: {{ task }} Format: {{ output_format }} ``` ## Frequently Asked Questions What does RTF stand for in prompt engineering? RTF stands for Role, Task, Format. You assign the AI a specific expert persona (Role), tell it exactly what to do (Task), and specify how the output should be structured (Format). It is one of the most minimal structured frameworks — just three components — but those three components address the most impactful variables in any AI prompt. How is RTF different from TAG or APE? RTF explicitly includes a Role component, making it the simplest framework that combines persona assignment with task and output formatting. TAG (Task, Action, Goal) is outcome-focused but lacks a Role. APE (Action, Purpose, Expectation) is audience-focused but also lacks a Role. RTF is the easiest entry point for anyone who wants to use role prompting as part of their standard workflow. When should I use RTF over RACE? Use RTF when your prompt is clear and self-contained — you know the role, the task, and the format, and no additional context is needed. Use RACE when the task requires background information the AI wouldn't otherwise have (e.g., your product's specific brand voice, your audience's particular characteristics, or domain-specific constraints). RACE's Context component is the key differentiator. What belongs in the Format component of RTF? Format should specify the output's structure (bullet list, numbered steps, table, prose paragraphs, markdown), its length (word count, number of items, sections), its tone or register (formal, casual, technical), and any hard constraints (no jargon, must include a CTA, max 150 words). The more specific your Format, the more predictable and usable the AI's output will be. Can I use RTF as a template for repeated tasks? Absolutely — this is one of RTF's greatest strengths. Because it has only three components and clear semantics, RTF is ideal for creating reusable prompt templates. You can fix the Role and Format for recurring tasks (weekly reports, customer emails, code reviews) and only update the Task description each time you use the template. Does RTF work for coding tasks? Yes. Set Role to the type of developer needed (e.g., 'senior Python engineer'), Task to the specific coding job (e.g., 'refactor the following function to use async/await'), and Format to the output structure (e.g., 'return only the refactored code followed by a brief explanation of each change, no markdown code fences'). RTF handles focused coding tasks well. Is RTF a good first prompt framework to learn? RTF is arguably the best starting framework for beginners. It is minimal enough to learn in under a minute, yet it addresses the three variables that most dramatically improve AI output quality: who is speaking (Role), what they are doing (Task), and what the result should look like (Format). Mastering RTF builds the intuitions needed for more complex frameworks like RACE or RISEN. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/scamper [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › SCAMPER Technique Creative Technique Intermediate # SCAMPER Technique Seven directed creativity lenses — Substitute, Combine, Adapt, Modify, Put to other uses, Eliminate, Reverse — that systematically break habitual thinking and generate ideas no open-ended brainstorm would surface. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Substitute · Combine · Adapt · Modify · Put to use · Eliminate · Reverse Best for Product innovation, process redesign, creative ideation Complexity Medium — works best with an existing concept to challenge Token cost Medium to High **TL;DR:** SCAMPER (Substitute, Combine, Adapt, Modify, Put to other uses, Eliminate, Reverse) applies seven directed creativity lenses to an existing product or process, breaking habitual thinking and reliably surfacing ideas that open-ended brainstorming misses. Use it for product innovation, process redesign, and any ideation challenge where a concrete starting concept already exists. Select 2-3 lenses most relevant to your challenge rather than forcing all seven, and ask for 2-3 specific ideas per lens with a brief feasibility rating. ## What is the SCAMPER Technique? SCAMPER is a structured creative thinking technique that helps you systematically generate new ideas by applying seven distinct lenses to an existing product, process, or concept. The acronym stands for **Substitute**, **Combine**, **Adapt**, **Modify** (or Magnify/Minify), **Put to other uses**, **Eliminate**, and **Reverse** (or Rearrange). Originally developed by educator Bob Eberle in the 1970s, SCAMPER has become a staple of design thinking, product management, and innovation workshops worldwide. In the context of AI prompting, SCAMPER is particularly powerful because language models can rapidly work through each lens, generating dozens of ideas in seconds. Rather than asking an AI to "brainstorm product ideas" and receiving generic suggestions, a SCAMPER prompt directs the model to interrogate specific assumptions — what can be replaced, what can be merged, what can be removed — producing far more original and actionable output. The technique requires a starting point: an existing product, service, process, or idea that you want to reimagine. This constraint is a feature, not a limitation. The best creative ideas often emerge from disciplined questioning of what already exists, not from unconstrained ideation. ## When to Use SCAMPER 🛠️ ### Product Innovation Challenge every assumption of an existing product — its materials, form factor, features, and target use cases — to identify opportunities for meaningful differentiation. 🔄 ### Process Redesign Apply Eliminate and Reverse lenses to business workflows to strip away unnecessary steps and question whether the sequence of operations is optimal. 🎨 ### Creative Brainstorming Use SCAMPER as a structured alternative to open-ended brainstorming when a team is stuck in conventional thinking or producing derivative ideas. 📚 ### Curriculum & Content Design Reimagine how educational content or media is structured — what could be adapted from another discipline, combined with interactive elements, or eliminated to reduce complexity. 💼 ### Business Model Innovation Question core business model assumptions: substitute the revenue stream, combine the offering with a platform, or reverse the value chain to find disruptive positioning. 🖥️ ### UX & Feature Ideation Apply SCAMPER to a specific user flow or feature to generate UX improvement ideas that go beyond incremental refinement and surface genuinely novel interaction patterns. ## How to Use SCAMPER in a Prompt 1. 1 ### Define the subject and set the role Start by specifying what you are applying SCAMPER to — the existing product, process, or concept. Give the AI a relevant persona such as a product strategist, innovation consultant, or UX researcher. A well-defined subject is essential because SCAMPER only works when there is something concrete to interrogate. Vague subjects produce vague ideas. 2. 2 ### Select your lenses Decide whether you want the AI to work through all seven SCAMPER lenses or a targeted subset. For a full brainstorm, use all seven. For a focused session — say, when you suspect over-engineering — use just Eliminate and Reverse. Fewer lenses means deeper exploration of each one. Specify your selection explicitly in the prompt. 3. 3 ### Set the idea quantity and constraints Instruct the AI how many ideas to generate per lens (two to three is a good starting point) and any domain constraints to respect — budget, technology maturity, regulatory environment, target audience. Constraints sharpen creativity rather than limiting it; they prevent the AI from generating ideas that are imaginative but impractical for your context. 4. 4 ### Specify the output format Define how you want the ideas presented: structured list with bold headers, markdown table with feasibility ratings, numbered paragraphs, or a one-line concept per lens. A clear format makes the output immediately usable in a workshop or design review. Ask for brief rationale alongside each idea so you can evaluate it without follow-up questions. ## Prompt Examples Full SCAMPER session — Paper Notebook ``` You are a creative product strategist using the SCAMPER technique to generate innovation ideas. Apply SCAMPER to the following product: a traditional paper notebook. Work through each lens in order: - Substitute: What materials, components, or processes could be replaced? - Combine: What could be merged with the notebook to add value? - Adapt: What ideas from other industries could be borrowed and applied? - Modify / Magnify: What could be exaggerated, enlarged, or altered? - Put to other uses: How could a notebook be used in completely unexpected ways? - Eliminate: What features could be removed to create a simpler or more focused product? - Reverse / Rearrange: What if the notebook worked backwards or the user's role was reversed? For each lens, generate 2–3 specific, actionable ideas. Format as a structured list with bold section headers. ``` Targeted lenses — Gym Membership Business Model ``` You are an innovation consultant facilitating a SCAMPER brainstorming session. Product under review: a standard gym membership model. Apply only the Substitute, Combine, and Reverse lenses to challenge the existing business model. For each lens: 1. Describe the core assumption being challenged 2. Propose two concrete alternative concepts 3. Rate each concept's feasibility (Low / Medium / High) and novelty (Low / Medium / High) Keep each idea concise — one to two sentences. Present results in a markdown table. ``` Education use case — Fraction Lesson Redesign ``` You are a curriculum designer helping educators redesign lesson plans using SCAMPER. Teaching topic: Introduction to fractions for Grade 4 students. Use the Adapt and Eliminate lenses specifically: - Adapt: What teaching methods from other subjects or disciplines could be borrowed to make fractions more intuitive? - Eliminate: What traditional elements of fraction instruction could be removed to reduce confusion and cognitive load? For each lens provide three ideas. For each idea, write: the idea itself, why it works pedagogically, and one potential drawback. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Breaks cognitive fixedness that stifles conventional brainstorming | Requires an existing subject — not suited for greenfield ideation from scratch | | Each lens is a reliable trigger for a distinct class of ideas | All seven lenses at once can produce an overwhelming volume of output | | Scales from quick individual use to full team innovation workshops | Ideas still need human evaluation; AI may generate impractical suggestions | | Works across virtually any domain — product, service, content, process | Less structured than design frameworks like JTBD or Lean Canvas for validation | | AI can generate a full SCAMPER analysis in seconds, far faster than human facilitation | | ## SCAMPER Technique Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` You are an innovation facilitator. Apply SCAMPER to: {{ product_or_process }} For each lens, generate {{ ideas_per_lens }} concrete ideas: Substitute: What could be replaced? Combine: What could be merged? Adapt: What could be borrowed from elsewhere? Modify: What could be enlarged, reduced or changed? Put to other uses: Where else could this be used? Eliminate: What could be removed? Reverse: What could be reordered or flipped? Finish by ranking the top 3 ideas by impact and effort. ``` ## Prompt Flows That Use SCAMPER Technique SCAMPER Technique is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The UX Strategist](https://www.promptedit.app/prompt-flow/the-ux-strategist) · Product Design ## Frequently Asked Questions What does SCAMPER stand for? SCAMPER is an acronym for Substitute, Combine, Adapt, Modify (or Magnify/Minify), Put to other uses, Eliminate, and Reverse (or Rearrange). Each letter represents a creative lens you apply to an existing idea, product, or problem to generate new possibilities. The technique was developed by Bob Eberle, building on Alex Osborn's earlier brainstorming work. How do I use SCAMPER in a prompt? Tell the AI to act as a creative strategist or innovation consultant, specify the product, service, or problem you want to rethink, and then ask it to work through each SCAMPER lens systematically. You can apply all seven lenses at once for a comprehensive brainstorm, or pick just two or three that are most relevant to your challenge. Is SCAMPER only useful for product design? No. SCAMPER works for any situation where you want to break out of habitual thinking. It is equally effective for redesigning business processes, improving lesson plans, refreshing marketing campaigns, rethinking user experiences, and even solving personal problems. The key is that there must be an existing concept to interrogate. Which SCAMPER lens is most powerful? It depends on your goal. Substitute tends to produce practical, near-term improvements. Combine often yields the most distinctive new concepts. Reverse/Rearrange is the most disruptive — it forces you to question fundamental assumptions. For radical innovation, start with Reverse. For incremental improvement, start with Substitute or Eliminate. Can I apply SCAMPER to only some of the lenses? Absolutely. Using all seven lenses in one session is useful for deep brainstorming, but in many situations you will get better-quality ideas by selecting two or three lenses that best fit your challenge. For example, use Eliminate + Reverse when you suspect a process is over-engineered, or Combine + Adapt when you want to create something genuinely novel. How is SCAMPER different from simple brainstorming? Standard brainstorming is open-ended and often produces ideas that cluster around familiar patterns. SCAMPER is a directed creativity technique — each lens acts as a constraint that forces thinking in a specific direction. This structure helps break cognitive fixedness and reliably surfaces ideas that free brainstorming would miss. Can SCAMPER be combined with other prompt frameworks? Yes. A common combination is to pair SCAMPER with Chain-of-Thought prompting, asking the AI to reason through each lens step by step before generating ideas. You can also wrap SCAMPER inside a RACE or RISEN structure to add Role, Context, and output Expectations for more focused results. ## Origin & Sources 1. Wikipedia (2026). [SCAMPER](https://en.wikipedia.org/wiki/SCAMPER). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/seed [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › SEED Framework Structured Prompting Beginner # SEED Framework Plant a vivid scene, explain your goal, show an example of the style, and define what is off-limits — giving AI just enough creative direction to produce imaginative, on-brand content every time. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Set the scene, Explain, Examples, Define Best for Creative writing, brand content, constrained generation Token cost Low **TL;DR:** SEED (Set the scene, Explain the goal, Examples, Define constraints) is a creative prompt framework that simultaneously gives the AI imaginative direction through vivid scene-setting and firm guardrails through explicit constraints — solving both the "too generic" and "too restricted" failure modes. Use it for literary content, brand copywriting, thought leadership, and any creative task requiring both voice distinctiveness and boundary compliance. Even a single short example in the Examples component dramatically shifts the quality and style of the output. ## What is the SEED Framework? The SEED framework is a creative-first prompt structure that gives AI models both the imaginative direction and the necessary constraints to produce compelling, on-brand content. Its four components — **Set the scene**, **Explain the goal**, **Examples**, and **Define constraints** — work together to give the AI a vivid creative context, a clear objective, a stylistic reference, and the boundaries within which to work. - **S — Set the scene vividly** - **E — Explain the goal precisely** - **E — Provide a style-defining example** - **D — Define constraints explicitly** The core insight behind SEED is that creative prompts often fail in one of two ways: they are either too open (producing generic, aimless output) or too rigid (producing mechanical content that lacks voice). SEED solves both problems simultaneously. Scene-setting and examples provide creative direction; the constraints component provides the guardrails that protect brand safety, legal requirements, or stylistic standards. - **Set the scene:** Establish the creative world, atmosphere, or scenario the AI should inhabit. - **Explain the goal:** Define the specific content deliverable and what it should achieve. - **Examples:** Provide stylistic references, tone samples, or structural models. - **Define constraints:** Set the non-negotiable boundaries of content, style, and format. ## When to Use the SEED Framework ✍️ ### Literary and Creative Fiction Establish the fictional world and atmosphere in Set the scene, define the narrative goal, provide a prose style reference, and constrain the output's scope and structure. 🛍️ ### Brand and Product Copywriting Immerse the AI in the brand world through scene-setting, specify the copy goal, share brand voice examples, and restrict terms, superlatives, or claims that are off-brand or legally sensitive. 📰 ### Thought Leadership and Opinion Pieces Set the editorial voice and publication context, explain the argument to be made, provide a paragraph in the target style, and define structural constraints like no listicles or no more than 700 words. 🎮 ### Game and Narrative Design Describe the game world's lore and aesthetic, define the narrative deliverable (quest description, NPC dialogue, item lore), share existing in-world text as style examples, and constrain tone and vocabulary to match the game's established voice. 🎨 ### Image and Visual Prompting Set the visual world, explain the creative intent, reference art styles or artists as examples, and define visual constraints like color palette, composition rules, or content restrictions. 📧 ### Campaign and Social Content Establish the campaign theme and emotional tone, define each content piece, share sample posts in the desired style, and constrain word counts, hashtag use, and prohibited messaging. ## How to Use the SEED Framework 1. 1 ### Set the scene vividly Do not describe the scene abstractly — make it sensory and specific. Use evocative details: the era, the atmosphere, the emotional register, the world the content inhabits. The scene-setting component is what transforms an AI response from generic to genuinely immersed in the right creative context. Think of it as writing the creative brief that an agency would need before starting a campaign. 2. 2 ### Explain the goal precisely State the specific deliverable and the effect it should have on the reader. A good goal statement includes: what to create, who it is for, and what the reader should feel, know, or do after consuming it. Goals like "write something engaging" are too vague; "write an opening paragraph that makes a busy founder stop scrolling and read on" is specific enough to guide creative decisions. 3. 3 ### Provide a style-defining example Even a single sentence or short paragraph can calibrate the AI's voice and style dramatically. Choose an example that demonstrates the exact register, rhythm, and vocabulary you want. If you have previous content you are happy with, paste a representative excerpt. If you are starting fresh, find a published piece whose style you want to emulate and quote a few lines. 4. 4 ### Define constraints explicitly List every boundary the output must respect. Think in categories: content constraints (what topics are off-limits), structural constraints (word count, number of sections), stylistic constraints (words or phrases to avoid, sentence length norms), and tonal constraints (not sarcastic, not promotional, not technical). Explicit constraints prevent the most common creative prompt failures and save revision cycles. ## Prompt Examples Literary Short Story Opening ``` Set the scene: A quiet coastal town in 1950s Ireland. The atmosphere is melancholy but not hopeless — faded fishing boats, the smell of salt and peat smoke, a community bound together by shared hardship and quiet pride. Explain the goal: Write the opening page of a literary short story centered on an elderly fisherman who has not gone to sea in three years following an accident. Today he walks to the harbor for the first time. Examples: The prose style should echo the quiet observation of William Trevor or John McGahern — short declarative sentences, close third-person perspective, emotion conveyed through action and detail rather than stated directly. Define constraints: Maximum 300 words. No flashbacks in this opening page. Do not reveal the nature of the accident yet. The final sentence should be the man looking at the sea, not yet stepping toward it. ``` Premium Brand Product Descriptions ``` Set the scene: We are launching a premium direct-to-consumer olive oil brand called Stavros & Sons. The brand world is modern Mediterranean — sun-bleached stone walls, terracotta, the unhurried rhythms of a Peloponnese village. The tone is warm, knowledgeable, and quietly proud, never flashy or corporate. Explain the goal: Write three product descriptions for our website — one for our Extra Virgin Olive Oil, one for our Chili-Infused Olive Oil, and one for our Herb Collection gift set. Examples: "Pressed from hand-picked Koroneiki olives within hours of harvest, this oil carries the grassy, peppery finish that only early-season fruit can deliver. It is not for cooking. It is for the bread, the salad, the moment." Define constraints: Each description should be 60-80 words. No ingredient lists (those appear separately). Do not use the words luxury, artisan, or premium. Avoid superlatives like best or finest. ``` Startup Blog Thought-Leadership Article ``` Set the scene: A SaaS startup blog aimed at founders and operators of early-stage companies. The publication voice is direct, practical, and occasionally dry — like getting advice from a smart friend who has already made the expensive mistakes. Explain the goal: Write a thought-leadership article arguing that most early-stage startups delay pricing their product too long and that this indecision costs them more than any pricing mistake would. Examples: "Most founders treat pricing like it is a final exam — something to study for, prepare perfectly, then sit. In practice, pricing is more like a first draft. The only way to make it good is to ship it, learn from the response, and revise." Define constraints: 600-800 words. First-person plural (we) voice. No numbered lists — this should read as a cohesive argument, not a listicle. No references to specific companies. End with one concrete action the reader can take today. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Balances creative freedom with clear, explicit guardrails | Scene-setting requires creative effort from the prompter | | Scene-setting produces more immersive, contextually rich output | Less suitable for analytical or instructional tasks | | Works for both fiction and brand content generation | Output quality heavily depends on example quality | | Constraints component reduces costly revision cycles | Not designed for iterative or multi-step reasoning workflows | ## SEED Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Scene: {{ scene }} Explain the goal: {{ goal }} Example: {{ style_example }} Define constraints: {{ constraints }} ``` ## Prompt Flows That Use SEED Framework SEED Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The UX Strategist](https://www.promptedit.app/prompt-flow/the-ux-strategist) · Product Design ## Frequently Asked Questions What does SEED stand for in prompt engineering? SEED stands for Set the scene, Explain the goal, Examples, and Define constraints. It is a structured prompt framework designed to balance creative freedom with clear boundaries. By establishing the scenario, clarifying the objective, providing stylistic examples, and defining what is off-limits, SEED produces creative and generative content that is imaginative yet bounded. How does the Examples component in SEED work? In SEED, the Examples component provides stylistic or structural references that show the AI what kind of output you want. Unlike frameworks where examples are used to demonstrate reasoning, SEED examples are about creative direction: a writing sample in the style you want, a structure you admire, or a tone reference. Even a single short example can dramatically shift the character of the AI's output. When should I use SEED over other creative frameworks? Use SEED when you need creative content but also need guardrails. If you want a story but have brand safety requirements, SEED's Define constraints component handles that. If you want generated content but need it to stay within a specific stylistic or structural range, SEED's scene-setting and examples give the AI creative direction without complete freedom. SEED is the creative framework with the strongest constraint vocabulary. What makes a good scene-setting in SEED? Good scene-setting is vivid and specific. It paints a picture of the world, context, or scenario the AI should work within. For a short story, it might describe the setting, era, and emotional atmosphere. For a marketing piece, it might describe the brand world and campaign theme. The goal is to immerse the AI in the creative context before any generation happens. Can SEED be used for non-fiction content? Yes. While SEED is most naturally suited to creative and generative content, it works well for non-fiction that requires imagination: thought leadership articles with a strong point of view, opinion pieces, brand storytelling, and narrative case studies. The framework is particularly useful when non-fiction needs a distinct voice or creative angle rather than a purely informational approach. How specific should the Define constraints component be? Be as specific as the project requires. Constraints can cover: content boundaries (no violence, no competitor mentions, no medical claims), structural limits (maximum 500 words, exactly three sections), stylistic exclusions (no cliches, no passive voice, no rhetorical questions), and tonal boundaries (not sarcastic, not preachy, not overly technical). The constraints component protects the output while leaving room for creativity. Is SEED suitable for AI image prompting as well as text? Yes. SEED translates naturally to image generation prompts. Set the scene establishes the visual world. Explain the goal defines what the image should communicate or achieve. Examples can reference art styles, artists, or existing images. Define constraints sets visual limits: color palette restrictions, content boundaries, composition rules. The framework works across any creative generation modality. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/self-consistency [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Self-Consistency Prompting Reasoning Intermediate # Self-Consistency Prompting Generate multiple independent reasoning paths for the same question, then aggregate the most consistent answer — dramatically increasing accuracy on tasks where a single chain of thought can go wrong. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Wang et al., Google Brain (2022) Builds on Chain-of-Thought Prompting Mechanism Majority vote across N reasoning paths Typical paths 5–20 per query **TL;DR:** Self-Consistency (Wang et al., 2022) generates multiple independent Chain-of-Thought reasoning paths for the same question, then selects the most common final answer by majority vote — improving accuracy by 10-18% on reasoning benchmarks. Use it for high-stakes math, logic, and scientific reasoning where a single reasoning chain may make a compounding error. Run 5-10 paths with temperature 0.5-0.8 to ensure diversity, then aggregate answers programmatically or with a follow-up aggregation prompt. ## What is Self-Consistency Prompting? Self-Consistency is a decoding strategy and prompting technique introduced by Wang et al. at Google Brain in 2022. It extends Chain-of-Thought prompting by sampling multiple diverse reasoning paths for the same question and aggregating their final answers via majority vote. The core intuition: if a correct answer can be reached through many different valid reasoning chains, then generating multiple chains and picking the most common final answer is far more reliable than trusting any single chain. Reasoning errors that lead to wrong answers tend to vary between paths, while correct reasoning converges on the same answer from different angles. The process has three steps: - **Sample:** Run the same Chain-of-Thought prompt N times with temperature > 0 to get N diverse reasoning paths. - **Extract:** Parse the final answer from each reasoning path. - **Aggregate:** Select the answer that appears most frequently (majority vote). In the original paper, Self-Consistency improved Chain-of-Thought accuracy on mathematical and commonsense reasoning benchmarks by 10–18 percentage points. ## When to Use Self-Consistency 🧮 ### Math Word Problems Multi-step arithmetic and algebra where calculation errors in any single chain can compound to a wrong final answer. ⚖️ ### Logical Reasoning Syllogisms, deductive reasoning, and multi-premise logical problems where a single wrong inference step ruins the conclusion. 📋 ### Multiple-Choice Questions High-stakes multiple-choice tests (bar exams, medical licensing, standardized tests) where confidence and accuracy are critical. 🔬 ### Scientific Analysis Interpreting experimental data, reasoning about causality, or evaluating competing hypotheses where the "right" answer matters. 💼 ### High-Stakes Decisions Business decisions, risk assessments, and strategic choices where a reasoning error could have significant consequences. 🏥 ### Medical Diagnosis Support Differential diagnosis reasoning where ruling out wrong conclusions before committing to a recommendation is essential. ## How to Use Self-Consistency 1. 1 ### Write a strong Chain-of-Thought prompt Self-Consistency amplifies CoT — start with a well-crafted CoT prompt (zero-shot with "Let's think step by step" or few-shot with worked examples). The diversity of reasoning paths depends on the quality of the base prompt. 2. 2 ### Set temperature above 0 Use temperature 0.5–0.8 when calling the API. Temperature 0 (greedy decoding) produces identical outputs every time — you need diversity to benefit from self-consistency. Each run should take a meaningfully different reasoning path. 3. 3 ### Run the prompt N times Sample 5–20 independent responses for the same query. More paths = higher confidence, but also higher cost. For most practical tasks, 5–10 paths provide substantial improvement with manageable cost. 4. 4 ### Extract and aggregate final answers Parse the final answer from each response. Count occurrences and select the majority answer. For text answers, you may need a normalization step or an aggregation prompt: "Given these 5 answers, what is the most common conclusion?" ## Prompt Examples Self-Consistency Setup (run this prompt 5–10 times) ``` Solve this problem step by step. Show all your work clearly. State your final answer on the last line, preceded by "Answer:". Problem: A train leaves Station A at 9:00 AM traveling at 80 mph. Another train leaves Station B (240 miles away) at 10:00 AM traveling toward Station A at 60 mph. At what time will the two trains meet? Let's think step by step. ``` Single-Prompt Self-Consistency Simulation ``` Solve the following problem using 5 different reasoning approaches. For each approach, show the full reasoning chain and state the answer. After all 5, identify which answer appears most frequently and declare it the final answer. Problem: If a store has a 30% off sale, and then offers an additional 20% off the sale price, what is the total percentage discount from the original price? Approach 1 — Algebraic: [reason through algebraically] Approach 2 — Numerical example ($100 original price): [reason through with concrete numbers] Approach 3 — Percentage multiplication: [reason through as decimals] Approach 4 — Step-by-step discount: [reason through each discount sequentially] Approach 5 — Common mistake check: [deliberately test if 30+20=50% is wrong and why] Final Answer (majority vote): ``` Aggregation Prompt (after collecting N responses) ``` I asked an AI model the same math problem 8 times and got these answers: 42, 42, 38, 42, 45, 42, 38, 42 Count the occurrences of each unique answer and identify the majority answer. Then explain whether the minority answers suggest any plausible alternative interpretations of the problem. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Proven 10–18% accuracy improvement on reasoning benchmarks | N× the token cost and latency of a single query | | Works without any prompt changes — just run CoT multiple times | Requires aggregation logic — more complex to implement | | Naturally surfaces uncertainty (low consensus = low confidence) | Not useful for creative, open-ended, or subjective tasks | | Composable with CoT, Few-Shot, and Role Prompting | Majority vote can be wrong if errors are systematic (not random) | ## Self-Consistency Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` {{ question }} Solve this problem {{ number_of_paths }} times, each time using a different line of reasoning. Label them Path 1, Path 2, and so on, and end each path with "Result:". Then compare the results: - Which answer appears most often? - Where do the paths disagree, and why? Final answer: the majority result, with a one-sentence confidence note. ``` ## Prompt Flows That Use Self-Consistency Prompting Self-Consistency Prompting is one step in these 3 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Debate Judge](https://www.promptedit.app/prompt-flow/the-debate-judge) · Decision-Making - [The Refiner](https://www.promptedit.app/prompt-flow/the-refiner) · Content Writing - [The Test Strategist](https://www.promptedit.app/prompt-flow/the-test-strategist) · Software Development ## Frequently Asked Questions What is Self-Consistency prompting? Self-Consistency is a prompting technique introduced by Wang et al. (2022) that generates multiple independent reasoning paths for the same question and then aggregates the results, typically by selecting the most common (majority-vote) final answer. Instead of trusting a single chain of thought, it samples diverse reasoning paths and uses their collective output to improve accuracy — particularly on tasks where reasoning mistakes are possible. How does Self-Consistency improve accuracy? By generating multiple independent reasoning chains (typically 5–20), Self-Consistency reduces the impact of any single incorrect reasoning path. If 8 out of 10 reasoning paths arrive at the same answer through different routes, that answer is highly likely to be correct. The diversity of paths catches different types of reasoning errors that would go undetected in a single CoT trace. When should I use Self-Consistency prompting? Self-Consistency is best for high-accuracy reasoning tasks where errors are costly: math word problems, logical deduction, multiple-choice questions, scientific reasoning, and any task where you need high confidence in the correctness of the final answer. It is overkill for creative tasks, open-ended questions, and tasks without a single correct answer. How many reasoning paths should I generate? Research shows significant gains with 5–10 paths for most tasks, with diminishing returns beyond 20–40. For most practical applications, 5 paths provide a good balance between accuracy gain and token cost. If cost is a concern, start with 5; if you need maximum accuracy for a critical decision, use 20+. How do I aggregate Self-Consistency results? The standard approach is majority voting: run the same prompt N times (with temperature > 0 for diversity), extract the final answer from each response, and select the answer that appears most frequently. For numerical answers, this is straightforward. For text answers, you may need a normalization step or a separate aggregation prompt to identify the most common conclusion. What temperature should I use for Self-Consistency? Use temperature > 0 (typically 0.5–0.8) to ensure diverse reasoning paths. If you use temperature 0 (greedy decoding), all paths will be identical and self-consistency adds no value. The diversity of paths is what makes the technique work — you need the model to explore different reasoning routes, not repeat the same one. Is Self-Consistency compatible with other techniques? Yes — Self-Consistency is most powerful when combined with Chain-of-Thought. Each of the N paths uses CoT reasoning, producing N diverse reasoning chains. The combination (CoT + Self-Consistency) is one of the best-performing prompting approaches for mathematical and logical reasoning benchmarks. You can also combine it with Role Prompting to add expert personas to each reasoning path. ## Origin & Sources 1. Wang et al. (2022). [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://arxiv.org/abs/2203.11171) . arXiv / ICLR 2023. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/serp [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › SERP Prompting SEO Framework Intermediate # SERP Prompting Structure AI-generated content for Google AI Overviews, Bing Copilot, and Perplexity — optimized for featured snippets, FAQ extraction, and AI search citations. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Search Engine Response Prompting Best for SEO content, AI Overviews, knowledge bases Key format Definition-first, question H2s, FAQ blocks Target platforms Google SGE, Bing Copilot, Perplexity **TL;DR:** SERP Prompting structures AI-generated content for extraction by Google AI Overviews, Bing Copilot, and Perplexity — using definition-first paragraphs, question-phrased H2 headings, and FAQ blocks that match the patterns AI search engines prefer to cite. Use it for informational pages, how-to guides, and knowledge base content where search visibility is a goal alongside human readability. Always pair generated content with a matching FAQPage or HowTo JSON-LD schema request in the same prompt. ## What is SERP Prompting? SERP Prompting — Search Engine Response Prompting — is a content generation strategy that structures AI prompts to produce output optimized for modern Search Engine Result Pages and AI-generated search overviews. As Google SGE, Bing Copilot, and Perplexity increasingly surface AI-synthesized answers, the structure of your content determines whether it gets cited. The core insight is that AI search engines do not read content the way humans do — they extract semantically clear snippets. Content that leads with direct definitions, uses question-phrased headings, and contains short authoritative paragraphs is extracted and cited at dramatically higher rates than long-form prose buried in introductions. SERP Prompting targets three main extraction patterns: - **Featured Snippets:** 40–60 word direct answers immediately after a question-phrased H2. - **FAQ Rich Results:** Structured Q&A blocks that map directly to FAQPage JSON-LD schema. - **AI Overview Citations:** Factually dense, clearly attributed paragraphs that AI search engines quote verbatim. ## When to Use SERP Prompting 📖 ### Definition & Knowledge Pages Topic definitions, glossary entries, and explainer articles where readers (and AI search) expect an immediate, precise answer to "what is X?" 📋 ### How-To Guides Step-by-step tutorials structured with numbered lists that Google's HowTo schema and AI Overviews can extract and display as rich results. ❓ ### FAQ & Support Pages Customer support knowledge bases and FAQ sections where every question-answer pair doubles as a potential featured snippet or AI Overview source. 📝 ### Blog Posts Targeting AI Overviews Informational blog content where you want sections surfaced by Google's AI Overviews rather than just ranked on page one. 🏢 ### Product & Feature Descriptions Product pages with structured benefit lists and clear factual claims that AI comparisons and shopping overviews can extract accurately. 🔬 ### Research Summaries Summaries of studies or technical reports that Perplexity and other research-oriented AI search engines are likely to retrieve and cite. ## How to Use SERP Prompting 1. 1 ### Identify the target query and extraction format Determine the exact search query you want to rank for and whether it calls for a definition (use a definition-first paragraph), a process (use numbered steps), or a comparison (use a table or bullet list). Different query intents map to different extraction formats. 2. 2 ### Instruct the model on information architecture Explicitly specify the content structure in your prompt: "Open with a direct definition under 60 words. Use H2s phrased as questions. After each H2, provide a 2–3 sentence featured-snippet-ready answer." The model cannot guess your structural requirements — state them explicitly. 3. 3 ### Request a FAQ section with search-intent phrasing Ask the model to generate 5–8 FAQ entries phrased the way users actually search ("How do I...", "What is the difference between...", "Is X safe to..."). These map directly to FAQPage schema and are prime candidates for Google AI Overview citations. 4. 4 ### Generate matching structured data (JSON-LD) After generating the content, prompt the model to output FAQPage, HowTo, or Article JSON-LD schema markup that mirrors the content. This pairs machine-readable signals with your well-structured prose and maximizes rich-result eligibility. 5. 5 ### Review for E-E-A-T signals Google's quality guidelines value Experience, Expertise, Authoritativeness, and Trustworthiness. Review generated content to ensure it includes specific figures, cites sources where possible, avoids vague claims, and reads as authoritative rather than generic. ## Prompt Examples Full SERP-Optimized Article Brief ``` You are an expert content writer optimizing for AI search overviews and featured snippets. Write a comprehensive guide on "how to use a password manager" for the keyword: "how to use a password manager" Structure the content as follows: 1. Open with a direct 40–60 word definition paragraph answering the keyword directly. 2. Use H2 headings phrased as questions (e.g., "What is a password manager?", "How do I set up a password manager?"). 3. After each H2, provide a concise 2–3 sentence answer suitable for a featured snippet. 4. Include a numbered step-by-step section for setup. 5. Add a 5-question FAQ section at the end. 6. Keep paragraphs under 3 sentences for scannability. Target audience: non-technical users. Tone: clear, helpful, authoritative. ``` FAQ Block with Schema Markup ``` Generate an FAQ section optimized for Google's AI Overviews for the topic: "intermittent fasting benefits" Requirements: - 6 questions phrased exactly as users search them (question-style, not keyword-style) - Each answer: 50–80 words, factual, cites a specific mechanism or study finding - Answer format: direct claim first, then supporting detail - Include at least one answer structured as a numbered list - End each answer with a concise one-sentence takeaway Also output the FAQPage JSON-LD schema markup for all 6 questions. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Increases likelihood of AI Overview and featured snippet placement | Rigid structure can feel formulaic for narrative or opinion content | | Produces content with clear, scannable information architecture | AI search citation is probabilistic — no format guarantees placement | | Generates schema markup alongside content in one prompt | Requires knowledge of target query intent to structure correctly | | Improves both human readability and machine extractability | Google's AI Overviews algorithm evolves rapidly, reducing predictability | ## SERP Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Write a section answering the search query: "{{ target_query }}" - Start with a 40–60 word definition-first answer that could be quoted on its own. - Use question-phrased H2 headings that match how people search. - Include a short list or table where the answer has steps or options. - Close with 3 FAQs and concise answers. Audience: {{ audience }} Primary source to cite: {{ source }} ``` ## Frequently Asked Questions What is SERP Prompting? SERP Prompting (Search Engine Response Prompting) is a strategy for structuring AI prompts so that the generated content is optimized for Search Engine Result Pages and AI-generated search overviews like Google SGE, Bing Copilot, and Perplexity. It focuses on producing clear definitions upfront, FAQ-style Q&A blocks, structured headings, and concise quotable answers that AI search engines prefer to surface. Why does content structure matter for AI search overviews? AI-generated search overviews (like Google's AI Overviews) pull snippets from web content using semantic extraction. They favor content with a direct definition in the first paragraph, clear H2 question headings, short factual paragraphs, and structured FAQ sections. Content that matches these patterns is significantly more likely to be cited or featured. How is SERP Prompting different from standard SEO content generation? Standard SEO prompting focuses on keyword density and article length. SERP Prompting specifically targets the information architecture that AI search engines extract from: definition-first paragraphs, question-based H2s, bulleted supporting evidence, and structured data-friendly layouts. It optimizes for AI citation rather than just human readership. What is a featured snippet and how do I target it? A featured snippet is the highlighted box Google shows at the top of search results that directly answers a query. To target it, your content should answer the query in 40–60 words immediately after the H2 that phrases the question, use a definition format ('X is...') or a numbered list for how-to queries, and avoid burying the answer in long introductory paragraphs. Should I use SERP Prompting for every piece of content? No. SERP Prompting works best for informational and knowledge-base content: definitions, tutorials, how-to guides, and FAQ pages. It is less appropriate for opinion pieces, creative writing, product landing pages, or narrative content where the structured format would feel unnatural and harm readability. Can SERP Prompting help with Perplexity AI and Bing Copilot citations? Yes. Both Perplexity and Bing Copilot use retrieval-augmented generation (RAG) and prioritize content that is clearly structured, factually dense, and well-organized. Authoritative citations within your content, clear headings, and concise answers all increase the likelihood your content is retrieved and cited by these AI search engines. What role do structured data and schema markup play in SERP Prompting? Structured data (JSON-LD schema like FAQPage, HowTo, and Article) signals to search engines how to interpret your content. When you generate content with SERP Prompting, you should simultaneously generate the corresponding schema markup — particularly FAQPage schema for Q&A sections — to maximize machine-readability and eligibility for rich results. ## Origin & Sources 1. Aggarwal et al. (2023). [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735) . arXiv / KDD 2024. 2. Google Search Central (2026). [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/smart [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › SMART Prompting Goal Framework Beginner # SMART Prompting Apply the proven SMART goal framework to your AI prompts — making each request Specific, Measurable, Achievable, Relevant, and Time-bound — for project planning and objective definition that produces genuinely actionable output. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as SMART Goals Prompting Best for Project planning, OKRs, goal setting, task scoping Token cost Low **TL;DR:** SMART prompting applies the goal-setting criteria — Specific, Measurable, Achievable, Relevant, Time-bound — to AI prompt design, addressing the five most common reasons prompts produce vague or unusable plans. Use it for project planning, OKR writing, strategic planning, and any task requiring a concrete, actionable plan rather than a generic overview. Not all five elements are needed for every task — apply whichever criteria eliminate the most ambiguity for your specific request. ## What is SMART Prompting? SMART prompting adapts the widely used goal-setting framework — **Specific**, **Measurable**, **Achievable**, **Relevant**, and **Time-bound** — to the design of AI prompts. Originally developed for organizational goal setting, the SMART criteria translate naturally into prompt engineering: they address the five most common reasons why prompts produce vague, unusable, or misaligned outputs. SMART prompting is particularly powerful for project planning, OKR definition, strategic goal setting, and any task where the AI needs to produce a concrete plan of action rather than a general overview. By applying each SMART principle to your prompt, you ensure the AI has a precise scope, clear success criteria, realistic constraints, a relevant focus, and a temporal frame — the five ingredients of genuinely actionable output. - **Specific:** Define exactly what you need, for whom, in what context. No vague requests. - **Measurable:** Define what success looks like in concrete, verifiable terms. - **Achievable:** Scope the request to what the AI can realistically produce at the required quality in one response. - **Relevant:** Focus the AI on what matters for the actual goal. Exclude tangential requests that dilute the output. - **Time-bound:** Set the temporal frame — the planning horizon, the deadline, the target period, or the historical scope. ## When to Use SMART Prompting 📅 ### Project Planning Define project scope, team constraints, and deadline as SMART components to produce realistic project plans with phases, milestones, and risk registers that a project manager can actually use. 🎯 ### OKR and Goal Setting Use Measurable to enforce specific metrics in key results, Achievable to calibrate ambition to real-world baselines, and Time-bound to anchor goals to a specific quarter or year. 📊 ### Strategic Planning Scope a strategic plan with Specific context, define Relevant priorities to exclude noise, and set Time-bound planning horizons (12-month, 3-year, 5-year) to produce focused, actionable strategic documents. ✅ ### Task and Deliverable Scoping Define individual work deliverables with SMART precision — specifying what is in scope (Specific), how quality will be judged (Measurable), what resources are available (Achievable), and when it is due (Time-bound). 📈 ### Marketing and Campaign Planning Define campaign objectives with Specific audience and channel data, Measurable KPIs per channel, Achievable targets based on historical benchmarks, and Time-bound campaign windows. 🏃 ### Personal Productivity and Coaching Apply SMART to personal goal-setting prompts: specific habit or skill to develop, measurable progress indicators, achievable daily commitments, relevance to a larger life goal, and a time-bound review cycle. ## How to Use SMART Prompting 1. 1 ### Make it Specific Replace every vague element in your prompt with a concrete equivalent. "Write a marketing plan" becomes "Write a 90-day email marketing plan for a DTC skincare brand targeting women aged 25-40 in the United States." Include the who, what, for whom, and in what context. Specific prompts produce specific outputs; vague prompts produce generic templates. 2. 2 ### Define what Measurable success looks like Tell the AI what a successful output must include or achieve. For plans: required deliverables, milestones, and metrics. For documents: required sections, minimum examples, and quality criteria. For goal frameworks: specific number formats, required baseline references, and verification criteria. Measurable requirements give you a way to evaluate whether the output is actually useful before accepting it. 3. 3 ### Scope the request to be Achievable Consider what the AI can realistically produce at the required depth in a single response. Break large projects into achievable sub-prompts. Specify any real-world constraints the plan must respect: team size, budget, technology available, regulatory environment. Achievable scoping prevents the AI from producing aspirational but impractical recommendations that ignore your actual constraints. 4. 4 ### Keep it Relevant and Time-bound Relevant means excluding tangential requests that dilute the output. State the primary objective and ask the AI to stay focused on it. Time-bound means anchoring the output to a specific period: a planning horizon, a deadline, a fiscal year, or a historical window. Time-bound plans have a beginning and an end, making them actionable rather than perpetually open-ended. ## Prompt Examples 90-Day Sales Onboarding Plan ``` Specific: Create a 90-day onboarding plan for a new B2B SaaS Account Executive joining a 15-person sales team. The role involves managing mid-market accounts in the financial services vertical. The company uses Salesforce, Outreach, and Gong. Measurable: The plan must define clear milestones at Day 30, Day 60, and Day 90. Each milestone should include 3-5 measurable competencies or activities the new hire should have completed. Include a self-assessment checklist for each milestone. Achievable: Assume the new hire has 3+ years of B2B SaaS sales experience and does not need basic sales training. Focus on company-specific knowledge, product expertise, and pipeline building for this specific vertical. Relevant: The plan should prioritize the activities most directly linked to the new hire generating their first qualified opportunity by day 90. Secondary priorities are product certification and CRM hygiene. Time-bound: The plan covers the first 90 calendar days. Week 1 should be the most detailed, with daily activities. Weeks 2-4 should have weekly themes. Months 2-3 should have monthly objectives. ``` Q3 Content Marketing OKRs ``` Specific: Write three OKRs (Objectives and Key Results) for a content marketing team of 4 people at a B2B HR software company. The team's focus for Q3 is increasing organic search traffic and improving lead quality from content. Measurable: Each Objective must be qualitative and aspirational. Each Objective must have exactly 3 Key Results. Each Key Result must be a specific number or percentage that can be unambiguously measured at quarter end. Achievable: The team's current baseline: 45,000 monthly organic visitors, 3.2% content-to-MQL conversion rate, publishing 8 articles per month. OKRs should represent ambitious but realistic stretch targets for a 90-day period, not 2x growth targets that would require significantly more headcount. Relevant: All OKRs must tie directly to the two strategic priorities: organic traffic growth and lead quality. Avoid vanity metrics like social shares or email open rates unless directly linked to these priorities. Time-bound: OKRs are for Q3 (July 1 to September 30). Key Results should be end-of-quarter targets, not ongoing activities. ``` HRIS Migration Project Plan ``` Specific: Create a project plan for migrating a company's internal HR data from an on-premise legacy system to a cloud-based HRIS. The company has 800 employees, data spanning 12 years, and the legacy system stores data in a proprietary format that requires custom ETL work. Measurable: The plan must include: total project timeline in weeks, number and names of distinct project phases, key deliverables per phase, a risk register with at least 5 identified risks and their mitigation approaches, and a definition of "migration complete" that all stakeholders can agree to. Achievable: Assume a project team of 1 project manager, 2 internal IT staff, and one external HRIS implementation partner. The plan should be realistic for this team size — not a best-case scenario requiring unlimited resources. Relevant: The business driver is a regulatory compliance deadline requiring all HR records to be accessible via the new system by a specific date. The plan must make this deadline the governing constraint for all phases. Time-bound: The compliance deadline is 6 months from project kickoff. The plan must work backward from this fixed end date. Flag clearly if the scope is not achievable in 6 months with the given team, and recommend scope adjustments. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Familiar framework — most professionals already know SMART goals | Not all five elements are relevant for every prompt type | | Measurable component ensures outputs are evaluable and actionable | Can feel formulaic for creative or exploratory tasks | | Achievable scoping prevents unrealistic or unimplementable plans | Achievable scoping requires self-awareness about AI capabilities and limits | | Works across domains: business, personal productivity, technical planning | Less suited for open-ended analysis or creative generation without clear objectives | ## SMART Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Turn this goal into a SMART plan: {{ rough_goal }} Specific: {{ what_exactly }} Measurable: {{ success_metric }} Achievable: {{ resources_and_constraints }} Relevant: {{ why_it_matters }} Time-bound: {{ deadline }} Output: the SMART goal in one sentence, followed by milestones with dates. ``` ## Prompt Flows That Use SMART Prompting SMART Prompting is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Sprint Planner](https://www.promptedit.app/prompt-flow/the-sprint-planner) · Project Planning ## Frequently Asked Questions What is SMART prompting in AI? SMART prompting applies the classic goal-setting framework — Specific, Measurable, Achievable, Relevant, and Time-bound — to the design of AI prompts. Each principle improves a different dimension of prompt quality: Specific eliminates vagueness, Measurable defines what success looks like, Achievable keeps the request within realistic scope, Relevant focuses the AI on what matters, and Time-bound sets the temporal frame for planning or analysis. Do I need to include all five SMART elements in every prompt? No. SMART is a checklist, not a rigid template. For simple requests, one or two SMART elements may suffice. A prompt asking for a meeting agenda may only need Specific and Time-bound. A prompt for a strategic plan benefits from all five. The framework is most powerful for project planning, goal-setting, and objective-definition prompts where all five dimensions are genuinely relevant. How does the Measurable element work in prompting? Measurable means defining what a successful output looks like in concrete, verifiable terms. For a marketing plan, measurable might mean: the plan must include KPIs for each channel, projected reach figures, and a conversion rate target. For a content brief, measurable might mean: the output should include 3 target keywords, a word count target, and a readability score goal. Without measurability, it is difficult to evaluate whether the AI's output is actually useful. What makes a prompt Achievable in the SMART sense? Achievable means scoping the request to what the AI can realistically deliver in a single response. Asking an AI to build a complete go-to-market strategy for a new product in one prompt is probably not achievable at the required depth. Breaking it into achievable sub-tasks — competitive landscape, positioning, channel strategy, messaging — produces far better results. SMART prompting encourages you to scope each prompt to a realistic unit of work. How is SMART prompting different from just being clear in your prompts? Clarity is necessary but not sufficient. You can write a clear prompt that lacks measurable success criteria (how will you know the output is good?), that is too broad to be achievable in one response, or that asks for analysis disconnected from the relevant time period. SMART prompting goes beyond clarity to address five specific dimensions that determine whether a prompt produces truly useful, actionable output. Is SMART prompting suitable for creative tasks? SMART can be applied to creative tasks, but some elements require interpretation. Specific defines the creative brief. Measurable might mean the output must include a specific structural element or hit a target word count. Achievable scopes the request to one creative piece rather than an entire campaign. Relevant ties the creative output to its intended use case. Time-bound might set the publication timeline or the era the creative piece should reference. Can SMART prompting help with AI-assisted project planning? SMART prompting is exceptionally well-suited to project planning tasks. Specific defines the project scope, Measurable sets the success metrics and deliverables, Achievable calibrates the plan to real-world constraints, Relevant focuses the plan on the project's core objectives, and Time-bound establishes the timeline and milestones. A SMART project planning prompt produces output that a project manager can actually use, not just a generic template. ## Origin & Sources 1. Wikipedia (2026). [SMART criteria](https://en.wikipedia.org/wiki/SMART_criteria). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/star [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › STAR Framework Narrative Framework Beginner # STAR Framework Four narrative building blocks — Situation, Task, Action, Result — that transform raw facts into compelling, credible stories for interviews, case studies, performance reviews, and anywhere achievement needs to be communicated clearly. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Situation · Task · Action · Result Best for Interview answers, case studies, retrospectives, storytelling Complexity Low — intuitive structure familiar from behavioral interviewing Token cost Low to Medium **TL;DR:** STAR (Situation, Task, Action, Result) is a narrative framework that transforms raw facts about an achievement into a structured story with a clear challenge, a defined responsibility, concrete steps taken, and a measurable outcome. Use it for behavioral interview answers, customer case studies, performance reviews, and portfolio write-ups where demonstrating competence through evidence is the goal. Always include a quantified Result — numbers are far more convincing than adjectives like "significantly improved." ## What is the STAR Framework? The STAR framework is a four-part narrative structure that organizes any achievement story into four clear components: **Situation** (the context and challenge), **Task** (the specific responsibility or goal), **Action** (the concrete steps taken), and **Result** (the measurable outcome). Originally developed as a behavioral interviewing technique, STAR has become one of the most widely used storytelling frameworks in professional communication. - **S — Situation:** Set the scene concisely - **T — Task:** Define the specific responsibility - **A — Action:** Detail every step taken - **R — Result:** Quantify the outcome In AI prompting, STAR serves two distinct purposes. First, you can feed the four components as structured input data and ask the AI to compose a polished narrative — interview answers, case studies, press releases, portfolio descriptions. Second, you can instruct the AI to produce its own responses in STAR format, ensuring structured, outcome-focused answers to open-ended questions. STAR is one of the most beginner-friendly prompt frameworks because its structure is inherently intuitive — every story worth telling has a beginning, a challenge, a response, and an ending. The framework simply makes this implicit structure explicit and ensures no critical element is left out. The most common mistake is skipping the Result, which is paradoxically the most important component — it is the proof that the Action actually worked. ## When to Use the STAR Framework 🎤 ### Interview Preparation Transform raw experiences into polished behavioral interview answers that demonstrate competence, self-awareness, and measurable impact — the three things hiring managers look for most. 📊 ### Case Studies Structure customer success stories and business case studies with a narrative arc that makes the problem vivid, the solution credible, and the outcome compelling for prospects and stakeholders. 📋 ### Performance Reviews Help employees and managers articulate contributions in a structured format that clearly connects actions to outcomes, making reviews more objective and evidence-based. 🔄 ### Project Retrospectives Document what happened on a project, what the team set out to achieve, the decisions made, and the measurable results — creating institutional knowledge that informs future work. 🖊️ ### Portfolio Write-Ups Frame portfolio projects and past work with a narrative that goes beyond what was built to explain the problem context, design decisions, and the impact those decisions had on real users or business metrics. 📣 ### Executive Storytelling Help leaders and founders communicate strategy decisions, company milestones, and turnaround stories in a structure that is easy for boards, investors, and employees to follow and remember. ## How to Use the STAR Framework 1. 1 ### Situation — Set the scene concisely Describe the context: the setting, the time frame, and the problem or challenge that created the need for action. The Situation should be specific enough to be real and relatable, but brief enough not to overshadow the Action. Aim for one to three sentences. Avoid generic setups — "we were growing fast" is weaker than "we had doubled headcount in six months with no formal onboarding process." 2. 2 ### Task — Define the specific responsibility State clearly what you (or the subject of the story) were specifically accountable for within this situation. The Task is different from the Situation — it is your particular role or goal, not the general problem. "My task was to reduce customer churn by 15% within two quarters" is a strong Task statement: specific, time-bound, and tied to a measurable target. Weak Tasks are vague: "I needed to help improve things." 3. 3 ### Action — Detail every step taken This is the longest and most important component. Describe exactly what you did — the decisions made, the methods used, the obstacles overcome, and why you chose this approach over alternatives. Use active verbs. Be specific about your individual contribution, especially in team contexts. The Action is where competence is demonstrated, so resist the urge to summarize — this is where you show your thinking and your work. 4. 4 ### Result — Quantify the outcome End with the measurable impact. Numbers are far more convincing than adjectives: "reduced support tickets by 40%" beats "significantly improved customer satisfaction." Where quantitative data is not available, be specific about qualitative outcomes: stakeholder reactions, decisions unlocked, risks eliminated, or team confidence restored. Always connect the Result back to the Task — confirm that what was set out to achieve was actually achieved, or explain what was learned if it was not. ## Prompt Examples Behavioral Interview Answer — Turning Around a Failing Project ``` You are a professional career coach helping a job seeker prepare for behavioral interviews. Use the STAR framework to write a polished, first-person interview answer to the following question: Question: "Tell me about a time you turned around a failing project." Here are the raw facts to work from: Situation: Six weeks before a major product launch, user testing revealed critical usability failures in the core checkout flow. The timeline and a signed enterprise contract worth $400K were both at risk. Task: As product manager, I was responsible for redesigning the checkout experience, restoring team confidence, and hitting the original launch date — with no additional budget. Action: I ran three rapid design sprints over two weeks, brought in a UX contractor for one week, launched daily stakeholder updates, and negotiated with engineering to de-scope two non-critical features to protect the timeline. Result: We launched on time. Checkout completion improved 34% within the first month and the enterprise contract was signed two weeks post-launch. Write a polished 200–250 word first-person answer. Use natural, confident language — no bullet points. It should sound practiced but not scripted. ``` Customer Case Study — Logistics Software ROI Story ``` You are a B2B content writer specializing in technology company case studies. Use the STAR framework to write a customer success case study from the following raw facts: Situation: Meridian Logistics, a mid-sized freight company, was losing 12% of customer orders due to manual data entry errors in their dispatch system. Customer complaints had risen 40% year over year and three key accounts had issued formal warnings. Task: Meridian needed to eliminate dispatch data entry errors within 90 days or risk losing their largest account, worth $2.1M annually. Action: They implemented an automated data validation layer integrated with their existing transport management system, ran a two-week staff retraining program, and introduced a real-time error dashboard for operations managers. Result: Error rate dropped from 12% to 0.4% within 60 days. Customer complaint volume fell 78%. The at-risk account renewed for a three-year term and increased their contract value by 15%. Format this as a 300-word case study with four clearly labeled sections (Situation, Challenge, Solution, Results). Use the company name throughout. Write in third person. Professional and factual tone. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Universally understood — familiar to HR, managers, and executives | Results can feel formulaic if not written with natural language variation | | Ensures every story has a measurable outcome, not just activity | Situation and Task are easy to conflate — requires careful distinction | | Beginner-friendly — intuitive even for people new to prompt engineering | Less suited to creative, exploratory, or analytical tasks | | Works for both input structuring and output formatting instructions | Strong results require quantifiable data — hard to use without real metrics | | Highly reusable for any professional communication involving achievement | | ## STAR Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Turn these notes into a STAR story for {{ purpose }}. Situation: {{ situation }} Task: {{ my_responsibility }} Action: {{ actions_taken }} Result: {{ measurable_result }} Keep it under {{ word_limit }} words and write in the first person. ``` ## Frequently Asked Questions What does STAR stand for in prompt engineering? STAR stands for Situation, Task, Action, and Result. You provide the AI with each of these narrative components and ask it to structure a story, case study, or account around them. The framework ensures the output has a clear setup, a defined challenge, a concrete response, and a measurable outcome — all the elements that make a narrative credible and compelling. Is STAR only useful for interview preparation? No. While STAR originated in behavioral interviewing, it is a highly versatile narrative structure for any task that involves telling a story about how a problem was solved. Use it for case studies, project retrospectives, performance reviews, portfolio write-ups, investor presentations, press releases, and any content that needs to communicate achievement or problem-solving credibility. What is the difference between Situation and Task in STAR? Situation describes the external context — the environment, the challenge that existed, the conditions at the time. Task describes the specific responsibility or goal that the person or team was accountable for within that situation. Situation answers 'what was happening?' and Task answers 'what were you specifically trying to achieve?' Keeping these distinct produces a much cleaner narrative than conflating them. How much detail should I put in the Action component? The Action component should be the most detailed section of a STAR narrative — it is where competence is demonstrated. Include specific steps taken, decisions made, who was involved, and any obstacles overcome. Vague actions like 'I worked with the team' are far less convincing than 'I facilitated three stakeholder alignment sessions and restructured the project timeline around two critical dependencies.' The more specific the Action, the more credible the story. How do I use STAR as a prompt structure vs. giving STAR content to the AI? You can use STAR in two ways. First, you can feed the four STAR components as input data and ask the AI to write a polished narrative — interview answers, case studies, press releases, portfolio descriptions. Second, you can instruct the AI to produce its own responses in STAR format, ensuring structured, outcome-focused answers to open-ended questions. Both approaches work well depending on whether you have the raw content or need the AI to generate it. Can STAR be used for technical writing and reports? Yes. STAR maps cleanly to incident reports (what happened, what we were trying to fix, what we did, what the outcome was), product launch retrospectives, engineering post-mortems, and UX research summaries. For technical audiences, you may want to relabel the sections — Problem, Objective, Approach, Outcome — which preserves the structure while using domain-appropriate language. What makes a strong Result in a STAR narrative? A strong Result is specific, measurable, and tied directly to the Action and Task. Quantify wherever possible: percentages, time saved, revenue impact, error rate reduction, user satisfaction scores. If hard numbers are not available, describe the qualitative outcome clearly: stakeholder response, team confidence, customer retention, reputation restored. A vague result like 'it went well' undermines the entire narrative. ## Origin & Sources 1. Wikipedia (2026). [Situation, task, action, result](https://en.wikipedia.org/wiki/Situation,_task,_action,_result). ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/step-back-prompting [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Step-Back Prompting Reasoning Beginner # Step-Back Prompting Before answering a specific question, ask the model to step back and name the underlying concept or principle. Then answer the question using that principle. Abstraction first, details second. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Zheng et al., Google DeepMind (2023) Also known as Abstraction prompting Reported gains Up to +27% on TimeQA (PaLM-2L) Token cost Medium (two stages) **TL;DR:** Step-Back Prompting (Zheng et al., Google DeepMind, 2023) splits reasoning into abstraction and application: first ask for the general principle behind a question, then answer the question using it. It improved results on physics, chemistry and multi-hop knowledge questions by 7–27% in the paper. Use it for questions where recalling the right rule or background is the hard part. ## What is Step-Back Prompting? Step-Back Prompting was introduced in the paper *Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models* by Huaixiu Steven Zheng and colleagues at Google DeepMind, published at ICLR 2024. The idea mirrors how experts approach unfamiliar problems: before calculating anything, a physicist asks which law applies, and a historian first establishes the timeline. The technique has two stages: - **Abstraction:** the model answers a higher-level "step-back" question about the concept, principle or background behind the original question. - **Reasoning:** the model answers the original question, grounded in the principle or facts it just retrieved. Why it works: specific questions often contain distracting details that pull the model towards a plausible but wrong path. Naming the principle first narrows the search to the right knowledge. In the paper, the approach improved PaLM-2L on MMLU Physics by 7%, MMLU Chemistry by 11%, TimeQA by 27% and MuSiQue by 7%. ## When to Use Step-Back Prompting ⚛️ ### STEM Questions Physics, chemistry and engineering problems where choosing the right law or formula is most of the work. 📚 ### Knowledge-Intensive QA Questions about people, events or organisations where the model should first recall the relevant history. 🔗 ### Multi-Hop Questions Questions that need several facts combined, where recalling the broader context first avoids missing a hop. 🛠️ ### Debugging and Root Cause Ask what class of problem this is (race condition, caching, encoding) before looking at the specific stack trace. ⚖️ ### Policy and Compliance Identify the governing rule or regulation first, then apply it to the specific case. 🎓 ### Teaching and Explanations Explaining the concept before the worked answer produces better tutoring responses. ## How to Use Step-Back Prompting 1. 1 ### Write the step-back question Ask about the principle, category or background behind the specific question. Keep it general enough that the answer applies beyond this one case. 2. 2 ### Let the model answer it first Either as a separate call or as a labelled "Principle" section in the same response. Ask for concise, factual output. 3. 3 ### Answer the original question using the principle Explicitly instruct the model to base its reasoning on the principle it stated, and to reason step by step if the problem is multi-step. 4. 4 ### Check for consistency If the final answer contradicts the stated principle, that is a strong signal of an error worth a second pass. ## Prompt Examples Physics Question ``` Question: What happens to the pressure of an ideal gas if the temperature is doubled and the volume is increased by a factor of 8? Step 1 (step back): What physics principle relates the pressure, temperature and volume of an ideal gas? State it as a formula. Step 2: Using that principle, answer the question step by step. Final answer on the last line. ``` Software Incident ``` Our checkout service returns stale prices for about 5 minutes after a price change, but only for logged-in users. First, step back: which general classes of problems cause data to be stale for a fixed period for one group of users but not another? List the top 3 with a one-line explanation each. Then, using those categories, rank the most likely causes in our case and propose one diagnostic check per cause. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Large reported gains on knowledge and STEM tasks (7–27%) | Extra stage adds tokens and latency | | Simple to learn and to add to existing prompts | Little benefit for simple lookups or writing tasks | | Stated principle makes errors easier to spot | A wrong principle in step 1 leads the whole answer astray | | Combines naturally with Chain-of-Thought and RAG | Choosing the right level of abstraction takes practice | ## Step-Back Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Question: {{ question }} Step 1 (step back): What general principle, concept or background knowledge is needed to answer this? {{ step_back_question }} Answer it concisely. Step 2: Using only that principle, answer the original question step by step. Final answer: one line. ``` ## Frequently Asked Questions What is Step-Back Prompting? Step-Back Prompting is a two-step reasoning technique from Google DeepMind (Zheng et al., 2023). First the model is asked a more general 'step-back' question about the concept or principle behind the problem. Then it answers the original question using that principle. Abstracting first helps the model retrieve the right knowledge before it gets lost in the specific details. How much does Step-Back Prompting improve accuracy? In the paper, Step-Back Prompting improved PaLM-2L performance on MMLU Physics by 7% and Chemistry by 11%, on TimeQA by 27% and on MuSiQue multi-hop questions by 7%. Gains were also reported with GPT-4 and Llama 2 70B. What is a good step-back question? A good step-back question asks about the general principle, category or history behind the specific question. For 'What happens to the pressure of an ideal gas if the temperature doubles and the volume increases eight times?' the step-back question is 'What physics principle relates pressure, temperature and volume?' (the ideal gas law). For 'Which school did Estella Leopold attend between August 1954 and November 1954?' it is 'What was Estella Leopold's education history?' What is the difference between Step-Back Prompting and Chain-of-Thought? Chain-of-Thought reasons forward through the specific problem step by step. Step-Back first moves up a level of abstraction to recall the governing concept, then reasons down to the answer. They combine well: step back to get the principle, then apply it with Chain-of-Thought. Can I do Step-Back Prompting in one prompt? Yes. Ask the model to first state the underlying principle or background in a labelled section, then answer the original question using it. The paper used separate abstraction and reasoning steps, but a single structured prompt works well in chat. When does Step-Back Prompting not help? It adds little for simple lookups or tasks with no underlying principle to recall, such as rewriting or formatting. It helps most on knowledge-intensive and multi-hop questions where the model needs to recall the right facts or rules first. ## Origin & Sources 1. Zheng et al. (Google DeepMind) (2023). [Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models](https://arxiv.org/abs/2310.06117) . arXiv / ICLR 2024. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/tag [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › TAG Framework Structured Framework Beginner # TAG Framework Three focused components — Task, Action, Goal — strip prompt writing to its essentials so you can get high-quality AI results fast, without over-engineering. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Components Task — Action — Goal Best for Quick, well-defined requests Complexity Very Low — the minimal viable prompt Token cost Low **TL;DR:** TAG (Task, Action, Goal) is the minimal viable prompt structure — three focused components that answer what needs doing, how to do it, and what success looks like — fast enough to use without friction on everyday requests. Use it for quick drafting, translation, list generation, and any well-defined task where context is self-evident and speed matters more than fine-grained control. The Goal component is the most commonly skipped — always include it to prevent the AI from optimizing for the wrong target. ## What is the TAG Framework? The TAG framework is a minimal three-part prompt structure built around three core questions: What needs to be done (**Task**)? How should it be done (**Action**)? What does success look like (**Goal**)? By answering these three questions explicitly in your prompt, you give the AI model everything it needs to produce a focused, useful response without unnecessary back-and-forth. - **T — Task:** Name what needs doing - **A — Action:** Give the specific instruction - **G — Goal:** Define the success criteria TAG is intentionally lean. Where frameworks like RACE or RISEN add components for persona, context, examples, and style, TAG strips things down to the smallest set of instructions that still produces reliable, on-target results. This makes it perfect for situations where you know the task well, the context is self-evident, and speed matters more than fine-grained control. Think of TAG as the quick-start prompt template you reach for when you have a clear job to be done and don't want to spend two minutes composing a structured brief. It is a great first framework to learn and a useful fallback even for experienced prompt engineers who need a fast answer. ## When to Use the TAG Framework ⚡ ### Fast Drafting When you need a first draft quickly — an email, a summary, a list of ideas — and the context is already clear from the subject matter itself. 🌐 ### Translation Tasks Specify what to translate (Task), any tone or style constraints (Action), and the target audience or publication context (Goal) for publication-ready translated text. 📝 ### Marketing Copy Define the copy type (Task), give creative direction like tone, word limits, or device (Action), and state the conversion or engagement goal (Goal). 📋 ### List Generation Ask for a list of ideas, options, or items (Task), specify how they should be generated or filtered (Action), and state the purpose they will serve (Goal). 🗂️ ### Reformatting Content Describe the source content and target format (Task), the transformation rules (Action), and how the reformatted content will be used (Goal). 🔄 ### Iterative Refinement When you have an existing draft and need targeted improvements — use TAG to specify the revision task, the improvement method, and the quality bar you are targeting. ## How to Use the TAG Framework 1. 1 ### Task — Name what needs doing State the high-level objective in one or two sentences. This is the category of work: "write a subject line," "translate this paragraph," "generate a list of blog post titles." Task sets the scope and the type of output expected. Keep it concrete and noun-focused. 2. 2 ### Action — Give the specific instruction Action is where you use strong, specific verbs to tell the AI exactly how to execute the task. This is where creativity direction, constraints, and methods live: "generate five options that create curiosity," "translate while preserving casual tone," "filter by relevance to small business owners." The more specific your Action, the less interpretation the model has to do. 3. 3 ### Goal — Define the success criteria Goal answers: what is this output for, and what makes it successful? Include the intended audience, the platform or use case, and any measurable quality criteria (under 50 characters, ready to publish, suitable for a non-technical reader). Goal prevents the AI from optimizing for a different target than the one you actually care about. ## Prompt Examples Email Subject Lines — Re-engagement Campaign ``` Task: Write a subject line for a re-engagement email campaign. Action: Generate five subject line options that create curiosity and a sense of urgency without being clickbait. Goal: Increase open rates among subscribers who have not opened an email in 90 days. Each subject line should be under 50 characters and work well on mobile. ``` Product Description Translation ``` Task: Translate the following product description from English to French. Action: Translate accurately while preserving the casual, friendly brand voice. Do not use overly formal French constructions. Goal: A ready-to-publish French product description for our e-commerce site targeting French-speaking customers in France and Belgium. --- Original text: "Meet Spark — the portable Bluetooth speaker that goes wherever you do. Waterproof, drop-proof, and loud enough to fill any room." ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Fastest structured framework to write — three short sections | No Role component — responses may lack the right expertise tone | | Easy to remember and apply without notes | No dedicated Context slot for nuanced background information | | Scales well for repetitive, templated tasks | Less effective for complex, multi-step, or creative tasks | | Good gateway framework for beginners | | ## TAG Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Task: {{ task }} Action: {{ action }} Goal: {{ success_criteria }} ``` ## Frequently Asked Questions What does TAG stand for in prompt engineering? TAG stands for Task, Action, Goal. Task defines what needs to be done at a high level, Action describes the specific steps or verbs the AI should apply, and Goal states the desired outcome or success criteria. Together these three elements create a minimal but effective prompt structure. When is TAG the right choice over a longer framework? TAG is best when your request is straightforward and you need a result quickly. If the task is well-defined, the context is obvious, and you do not need a specific persona or worked examples, TAG gets you there without the overhead of assembling four to six components. Use longer frameworks like RACE or RISEN when precision matters more than speed. How is TAG different from APE? Both are three-part beginner frameworks. TAG focuses on the what (Task), how (Action), and outcome (Goal). APE focuses on the what (Action), why/for whom (Purpose), and output quality (Expectation). TAG is more task-oriented and output-focused; APE builds in audience awareness more explicitly through the Purpose component. What is the difference between Task and Action in TAG? Task is the high-level objective — the overall category of what needs to be done (e.g., 'write a subject line', 'translate a document'). Action is the specific operational instruction — the verb-driven description of exactly how to accomplish it (e.g., 'generate five options that create urgency', 'translate while preserving casual tone'). Task sets scope; Action sets method. Can TAG handle creative writing tasks? Yes. Set Task to the type of creative output needed (short story, poem, tagline), Action to the specific creative direction (write in the style of noir fiction, use internal monologue, rhyme in ABAB), and Goal to the purpose or audience (engage readers aged 18-25, set a melancholy mood). Creative tasks benefit from a clear Goal to anchor the AI's choices. Should I add context even though TAG does not have a dedicated slot? Absolutely. TAG is a minimal framework, not a restrictive one. You can — and often should — add a brief context sentence within your Task or Action description. For complex tasks, consider upgrading to RACE or CARE which have dedicated Context slots. TAG is the starting point; adapt it as needed. Is TAG suitable for non-native English speakers learning prompt engineering? TAG is one of the most accessible frameworks for anyone new to prompt engineering, including non-native English speakers. The three components are short, concrete, and map directly to how most people naturally describe a task: what you need, how to do it, and what success looks like. It is an excellent first framework to practice. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/trace [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › TRACE Framework Structured Prompting Intermediate # TRACE Framework Define Task, Role, Audience, Context, and Expectation to produce content that speaks directly to specific readers — calibrated to their expertise, motivations, and decision-making criteria. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Also known as Task-Role-Audience-Context-Expectation Best for Marketing copy, audience-targeted content, communications Token cost Low **TL;DR:** TRACE (Task, Role, Audience, Context, Expectation) is a five-component framework whose dedicated Audience component produces genuinely reader-calibrated content by requiring you to define expertise level, motivations, vocabulary, and objections before the AI writes a word. Use it for marketing copy, internal communications, educational materials, and any content where the gap between generic and targeted output has measurable business consequences. Spend the most time on Audience — vague audience descriptions are the most common reason TRACE prompts underperform. ## What is the TRACE Framework? The TRACE framework is a five-component structured prompt technique that places the target audience at the heart of content creation. Its components — **Task**, **Role**, **Audience**, **Context**, and **Expectation** — guide the AI to produce content that is not just accurate and well-structured, but genuinely calibrated to the specific readers who will consume it. Most prompt frameworks tell the AI what to do and how to do it. TRACE adds a critical question: *who is this for?* The dedicated Audience component prompts you to define the reader's expertise level, motivations, vocabulary, and potential objections. This audience-first thinking is what separates content that resonates from content that merely informs. - **Task:** The specific content deliverable or writing assignment. - **Role:** The expert persona the AI should adopt to best serve the task and audience. - **Audience:** Who the content is for — expertise, demographics, motivations, and knowledge base. - **Context:** The situational background, topic details, and relevant facts. - **Expectation:** Output format, tone, length, and quality criteria. ## When to Use the TRACE Framework 📣 ### Marketing Copy Define the buyer persona in Audience to produce copy that addresses the right pain points, uses the right vocabulary, and converts the specific decision-maker you are targeting. 📰 ### Content Marketing and Blogging Specify reader expertise and search intent in Audience to calibrate depth of explanation, vocabulary choice, and assumed prior knowledge across blog posts and articles. 📢 ### Internal Communications Define the employee audience profile to produce change management communications, policy announcements, and leadership messages that address the specific concerns of frontline teams versus managers. 🎓 ### Educational Materials Specify the learner's starting knowledge level in Audience to produce training materials, e-learning scripts, and course content at exactly the right difficulty and pace. 💡 ### Product Documentation Differentiate between technical developer docs and end-user help articles by defining distinct audience profiles, ensuring each type of documentation assumes the appropriate level of technical knowledge. 🤝 ### Stakeholder Presentations Write presentation talking points and slides calibrated to investors, board members, customers, or technical teams by specifying their decision-making criteria and communication preferences in the Audience component. ## How to Use the TRACE Framework 1. 1 ### Define the Task State the specific content deliverable: a blog post introduction, a product announcement email, a one-page executive summary, or talking points for a presentation. The more specific the task definition, the more focused the output. Avoid vague tasks like "write some content about X." 2. 2 ### Assign the Role Choose an expert persona that matches both the content domain and the audience. A developer advocate writes differently about the same topic than a sales copywriter or a technical journalist. The role should reflect the perspective and communication style best suited to reach the specific audience. 3. 3 ### Profile the Audience Deeply This is TRACE's most important component. Define: job title and seniority, industry and company size, technical expertise level, primary motivations and goals, key concerns and objections, and vocabulary they use. The more specific your audience profile, the more targeted and effective the content. Generic audience descriptions produce generic content. 4. 4 ### Provide the Context and Expectation Context gives the factual background: the product, the situation, the data, the event. Expectation defines the output: format (email, article, bullet points), length, tone, and any structural requirements. Together these two components ensure the AI has both the subject matter and the delivery criteria it needs. ## Prompt Examples SaaS Feature Announcement Email ``` Task: Write a product feature announcement for our new AI-powered expense categorization feature. Role: You are a B2B SaaS product marketer with expertise in communicating technical features to non-technical buyers. Audience: CFOs and finance directors at mid-market companies (100-500 employees). They are not technical, they are time-poor, they care deeply about accuracy, compliance, and reducing their team's manual workload. They are skeptical of AI claims and respond to concrete ROI data. Context: Our expense management platform has added automatic AI categorization that reduces manual categorization time by an average of 73%. The feature learns from each company's historical categorization patterns. It also flags policy violations automatically. Expectation: A 200-word product announcement suitable for an email newsletter. Lead with the business outcome, not the technology. Include one specific statistic. Close with a clear next step. ``` Technical Blog Post Introduction for Non-Technical Readers ``` Task: Write a blog post introduction explaining what an API is. Role: You are a developer advocate who specializes in making technical concepts accessible to non-technical audiences. Audience: Small business owners and entrepreneurs who are evaluating software integrations for their company. They have no coding background, they are comfortable with business software, and they want to understand enough to have informed conversations with developers and make good purchasing decisions. Context: The post is part of a series called "Tech Basics for Business Leaders" on a B2B software company's blog. The target keyword is "what is an API for business". Expectation: A 150-word introduction that opens with a relatable business analogy, briefly defines the term without jargon, and previews what the reader will learn. Conversational tone. No code examples in the introduction. ``` Manager Communication: New Hybrid Work Policy ``` Task: Write talking points for a manager communicating a new hybrid work policy to their team. Role: You are an experienced HR communications specialist. Audience: A team of 12 software engineers at a tech company who have been fully remote for 3 years. They value autonomy, are skeptical of top-down mandates, and have varying personal situations (some with long commutes, some with young children). They will have questions about fairness and flexibility. Context: The company is introducing a policy requiring 2 in-office days per week starting next month. The manager has some flexibility to arrange which days work best for the team. Expectation: Five concise talking points for a team meeting. Each point should acknowledge the team's perspective before making the case. End with a list of three open questions the manager should invite the team to discuss. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dedicated Audience component produces genuinely reader-calibrated content | Requires more upfront effort than simpler three- or four-component frameworks | | Excellent for marketing, communications, and audience-specific writing | Audience profiling can be difficult if the reader base is undefined or diverse | | Five clear components with non-overlapping purposes | Less suited for purely analytical tasks where audience matters less than rigor | | Reduces the gap between generic AI output and targeted professional content | Five components can feel excessive for short, simple content requests | ## TRACE Framework Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Task: {{ task }} Role: You are {{ role }}. Audience: {{ audience_expertise_motivations_objections }} Context: {{ context }} Expectation: {{ expected_output }} ``` ## Prompt Flows That Use TRACE Framework TRACE Framework is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The Legacy Integrator](https://www.promptedit.app/prompt-flow/the-legacy-integrator) · Software Development ## Frequently Asked Questions What does TRACE stand for in prompt engineering? TRACE stands for Task, Role, Audience, Context, and Expectation. It is a five-component structured prompt framework that puts the target audience at the center of the prompt design. By explicitly defining who the content is for, TRACE ensures that AI output is calibrated to the right reading level, vocabulary, tone, and assumed knowledge base of the intended reader. What makes TRACE different from other structured frameworks? TRACE's distinguishing feature is its dedicated Audience component. While frameworks like RACE or RTF focus on role and task, TRACE explicitly asks you to define the characteristics of the reader: their expertise level, motivations, potential objections, and existing knowledge. This audience-first thinking produces content that resonates with specific people rather than a generic readership. When should I use the TRACE framework? Use TRACE when the audience profile significantly affects how the content should be written. Marketing copy for a technical developer audience reads very differently from copy for a non-technical executive. Blog posts for beginners require different vocabulary than whitepapers for specialists. TRACE is ideal for any content where getting the audience right is as important as getting the message right. How detailed should the Audience component be? The more specific the better. Include demographics (industry, job title, seniority), psychographics (goals, fears, motivations), knowledge level (expert, intermediate, novice), and any specific barriers or objections the reader might have. The Audience component is where most prompts fail — vague audience descriptions produce vague, unfocused content. Can TRACE be used for internal business documents? Yes. TRACE works well for any audience-specific document: investor updates for a board, technical specs for a development team, change management communications for frontline staff, or training materials for new employees. Defining the audience explicitly ensures the document assumes the right level of prior knowledge and addresses the right concerns. What is the difference between Context and Audience in TRACE? Audience describes the characteristics of the reader. Context provides the situational background the AI needs to understand the topic: the industry setting, the product being described, the event being communicated, or any relevant facts and constraints. Both are needed — Audience tells the AI who to write for; Context tells the AI what to write about. Is TRACE suitable for SEO content and blogging? Yes, TRACE is excellent for SEO and content marketing. Defining the Audience helps the AI calibrate vocabulary, reading level, and depth of explanation. The Task component can specify SEO goals (target keyword, search intent), and the Expectation component can define formatting requirements like header structure, word count, and meta description requirements. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/tree-of-thoughts [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Tree of Thoughts Reasoning Advanced # Tree of Thoughts (ToT) Explore multiple distinct reasoning paths simultaneously, evaluate each branch, and backtrack when needed — for solving complex problems where a single line of reasoning fails. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Origin Yao et al., Princeton & Google (2023) Builds on Chain-of-Thought Prompting Cost High (multiple model calls) Best for Planning, puzzles, strategic decisions **TL;DR:** Tree of Thoughts (Yao et al., 2023) extends Chain-of-Thought from a single linear path to a branching tree — the model generates multiple candidate reasoning steps, evaluates each branch, selects the most promising, and backtracks from dead ends, dramatically outperforming CoT on complex planning tasks. Use it for puzzles, strategic planning, software architecture decisions, and any problem where an early reasoning mistake irreversibly corrupts a linear chain. A simplified single-prompt version — "simulate 3 expert approaches, evaluate each, then solve with the best" — captures most of the benefit without multi-call orchestration. ## What is Tree of Thoughts Prompting? Tree of Thoughts (ToT) is an advanced reasoning framework introduced by Yao et al. in 2023. It extends Chain-of-Thought prompting from a single linear path to a tree structure of possible reasoning steps. At each node in the tree, the model generates multiple candidate "thoughts" (reasoning steps), evaluates their promise, and selectively expands the most viable ones — similar to how chess engines explore game trees. The framework recognizes a fundamental limitation of CoT: on hard problems, committing to a single reasoning chain early means that an early mistake propagates irreversibly to the final answer. ToT introduces deliberate exploration and self-evaluation, allowing the model to backtrack when a reasoning path leads to a dead end. The full ToT implementation requires multiple model calls orchestrated by a controller. However, a simplified "simulate multiple experts" approach can capture most of the benefit in a single prompt. ## When to Use Tree of Thoughts 🧩 ### Complex Puzzles Problems requiring exploration of multiple solution paths, like the Game of 24, logic puzzles, or constraint satisfaction problems. 🗺️ ### Strategic Planning Multi-step plans where early decisions constrain later options, such as project roadmaps, negotiation strategies, or product prioritization. 🎨 ### Creative Writing with Constraints Stories or content with structural requirements (plot arcs, metrical constraints) where multiple narrative approaches need evaluation. 🏗️ ### Software Architecture Evaluating multiple system design approaches before committing to an architecture with long-term consequences. 🔬 ### Research Hypothesis Generation Exploring multiple scientific hypotheses, experimental designs, or analytical frameworks before selecting the most promising. 📋 ### Decision Analysis Complex business decisions where multiple options must be evaluated against several criteria with trade-offs. ## How to Use Tree of Thoughts 1. 1 ### Define the problem clearly State the problem, constraints, and success criteria explicitly. ToT requires the model to evaluate branches — it needs clear criteria to judge which paths are promising. 2. 2 ### Generate multiple approaches Ask the model to generate 3–5 distinct high-level approaches to the problem without committing to any single one. These are the first-level branches of the tree. 3. 3 ### Evaluate and score each branch For each approach, ask the model to assess its viability: "Rate each approach on a scale of 1–5 for feasibility and quality. Explain why each may succeed or fail." 4. 4 ### Expand the best branches Select the 1–2 highest-scoring approaches and ask the model to elaborate on each, generating the next level of reasoning steps for those paths only. 5. 5 ### Commit and solve Select the winning path (by score or by inspection) and ask the model to produce the final, complete solution following that chosen reasoning branch. ## Prompt Examples Single-Prompt ToT (Simulate Multiple Experts) ``` You are facilitating a panel of 3 expert strategists. Each expert will propose a different approach to the problem, evaluate it honestly, and then the panel will vote on the best path. Problem: Our SaaS product has a 35% monthly churn rate. We have a $50k budget and 2 developers for the next quarter. Expert 1 (Product): Propose a product-focused solution. Expert 2 (Growth): Propose a growth/marketing-focused solution. Expert 3 (Customer Success): Propose a CS/retention-focused solution. For each expert: - Propose a specific action plan - Estimate expected churn reduction (%) - List the 2 biggest risks - Rate feasibility within our constraints (1–5) After all three, have the panel vote and explain the winning approach. ``` Architecture Decision with ToT ``` I need to design a real-time notification system for 100k concurrent users. Generate 3 distinct architectural approaches. For each: 1. Describe the architecture (2-3 sentences) 2. List the key technologies/components 3. Identify the primary bottleneck or risk 4. Rate scalability, implementation complexity, and cost (1-5 each) Then select the approach with the best balance for a 4-person engineering team at a Series A startup and explain the selection reasoning. ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Dramatically outperforms CoT on complex planning tasks | Full implementation requires multiple model calls — high cost | | Catches dead-end reasoning early through self-evaluation | More complex to implement and orchestrate than CoT | | Produces more robust solutions by exploring alternatives | Overkill for straightforward tasks where CoT suffices | | Can be simplified to work in a single prompt | Model self-evaluation quality determines branch selection accuracy | ## Tree of Thoughts Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` Problem: {{ problem }} 1. Propose {{ number_of_branches }} distinct approaches to this problem. Keep each to 2–3 sentences. 2. For each approach, evaluate: feasibility, risks, and likelihood of success (rate 1–10). 3. Discard the weakest approaches. Expand the best one or two into concrete next steps. 4. If an expanded branch hits a dead end, backtrack and explore the next-best approach. Output: the chosen approach, why it beat the alternatives, and the step-by-step solution. ``` ## Prompt Flows That Use Tree of Thoughts Tree of Thoughts is one step in these 3 multi-framework flows. Each flow chains frameworks in sequence for a real-world outcome. - [The Error Detective](https://www.promptedit.app/prompt-flow/the-error-detective) · Debugging - [The Legacy Integrator](https://www.promptedit.app/prompt-flow/the-legacy-integrator) · Software Development - [The Research Synthesizer](https://www.promptedit.app/prompt-flow/the-research-synthesizer) · Analysis & Research ## Frequently Asked Questions What is Tree of Thoughts prompting? Tree of Thoughts (ToT) is an advanced prompting framework that extends Chain-of-Thought reasoning by exploring multiple distinct reasoning paths simultaneously. Instead of a single linear chain, the model generates several candidate 'thoughts' or reasoning steps at each decision point, evaluates which paths are most promising, and backtracks when a path reaches a dead end — much like a search algorithm over a reasoning tree. How does Tree of Thoughts differ from Chain-of-Thought? Chain-of-Thought follows a single linear reasoning path from question to answer. Tree of Thoughts explores a branching tree of possibilities: at each reasoning step, the model generates multiple candidate next steps, evaluates their promise, selects the best ones, and continues. This allows the model to backtrack and explore alternatives, which CoT cannot do. When should I use Tree of Thoughts? ToT is best for problems where the correct approach isn't obvious upfront, where multiple valid solution strategies exist, and where dead ends need to be detected early. Classic examples: complex puzzles (Game of 24, crosswords), creative writing with structural constraints, strategic planning with dependencies, and multi-step code design. Is Tree of Thoughts expensive to run? Yes — ToT requires multiple model calls per reasoning step (one to generate candidates, one to evaluate them). This makes it significantly more expensive and slower than CoT. It is best reserved for problems where quality matters far more than cost or latency, or where the problem is genuinely intractable with simpler techniques. How do I implement Tree of Thoughts in a single prompt? A simplified ToT can be approximated in a single prompt with a 'simulate multiple experts' approach: ask the model to generate 3 different solution approaches, evaluate each one's strengths and weaknesses, select the best approach with justification, and then fully solve using that approach. This captures much of the benefit with fewer API calls. What search strategies are used in Tree of Thoughts? The original ToT paper explored two search strategies: Breadth-First Search (BFS), which evaluates all candidates at a given depth before going deeper, and Depth-First Search (DFS), which follows the most promising path as far as possible before backtracking. BFS is more exhaustive; DFS is more token-efficient. Can non-technical users benefit from Tree of Thoughts? Yes, through simplified prompts that ask the model to 'consider 3 different approaches before deciding'. You do not need to implement the full multi-call ToT framework to get significant benefit. The key insight — explore alternatives before committing — is applicable at any technical level. ## Origin & Sources 1. Yao et al. (2023). [Tree of Thoughts: Deliberate Problem Solving with Large Language Models](https://arxiv.org/abs/2305.10601) . arXiv / NeurIPS 2023. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-framework/zero-shot [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Zero-Shot Prompting Instruction-Based Beginner # Zero-Shot Prompting Give AI models clear, precise instructions without any examples — relying on the model's pre-trained knowledge to deliver accurate, well-formatted responses. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 Requires examples No — instructions only Best for Common tasks, rapid prototyping Key lever Prompt specificity and clarity Token cost Lowest (no example overhead) **TL;DR:** Zero-Shot Prompting gives the model a clear instruction without any worked examples, relying on its pre-trained instruction-following ability — the default way most people interact with AI and the right starting point before investing in more complex techniques. Use it for summarization, translation, drafting, Q&A, and any common task where the model's training is directly applicable. The highest-impact improvement is specificity: add explicit format, length, audience, and tone constraints to transform a vague request into a reliable, repeatable prompt. ## What is Zero-Shot Prompting? Zero-Shot Prompting is the most fundamental prompting style: you give the model a task or question with a clear instruction, and it responds based purely on its pre-trained knowledge and instruction-following training — no examples included. The "zero" refers to the number of demonstrations provided. Modern instruction-tuned models like GPT, Claude, and Gemini are extensively trained to follow zero-shot instructions reliably, which is why interacting with them "naturally" (asking a question or giving a command) already qualifies as zero-shot prompting. Zero-Shot Prompting is the right starting point for almost every task. Only escalate to Few-Shot or more complex techniques if the zero-shot result is inadequate — and the most impactful improvement is usually just making the instruction more specific. ## When to Use Zero-Shot Prompting ⚡ ### Rapid Prototyping Quickly explore whether a model can handle a task before investing in examples or fine-tuning. Zero-shot is the cheapest way to test feasibility. 📄 ### Summarization Condense articles, reports, or documents. Specify target length, audience, and format for best results. 💬 ### Q&A and Explanation Answer questions, explain concepts, or break down complex topics. Common knowledge tasks where the model's training is directly applicable. ✉️ ### Writing & Drafting Draft emails, blog intros, social posts, or meeting agendas with clear constraint-based instructions (tone, length, audience). 🔤 ### Translation Translate text between common languages — zero-shot is highly effective for standard language pairs. 🧹 ### Editing & Proofreading Grammar correction, style improvements, and tone adjustments work reliably as zero-shot instructions. ## How to Write Effective Zero-Shot Prompts 1. 1 ### State the task explicitly Begin with the core task verb. "Summarize", "Translate", "List", "Explain", "Rewrite", "Classify". Avoid vague openings like "Can you help me with..." — go straight to the request. 2. 2 ### Specify format, length, and structure Tell the model exactly what you want: "in 3 bullet points", "in under 100 words", "as a numbered list", "in JSON format". Without these constraints, the model uses defaults that may not fit your use case. 3. 3 ### Define the audience and tone Add context: "for a non-technical executive", "in a formal tone", "using simple language a 10-year-old could understand". This single addition dramatically improves relevance. 4. 4 ### Provide the input content clearly Separate your instructions from the content to process using clear markers like "Text:", "Article:", or triple backticks. This prevents the model from confusing your instructions with the content. ## Prompt Examples Weak vs Strong Zero-Shot ``` ❌ Weak: "Summarize this article." ✅ Strong: Summarize the following article in exactly 3 bullet points. Each bullet must be under 20 words and written for a busy executive who has no background in machine learning. Article: """ [article text here] """ ``` Structured Writing Request ``` Write a professional LinkedIn post announcing our new product launch. Constraints: - Tone: enthusiastic but not salesy - Length: 150–200 words - Include 3 key benefits as bullet points - End with a call-to-action encouraging readers to try the free trial - Do not use emojis Product: Prompt Edit — a free macOS app for managing AI prompt templates offline. ``` Constrained Analysis ``` Analyze the following customer feedback and provide: 1. The main complaint (1 sentence) 2. Underlying customer need (1 sentence) 3. Recommended action (1–2 sentences) 4. Urgency level: Low / Medium / High Feedback: "I've been trying to export my data for three days. The export button just spins forever. I have a client presentation tomorrow and I need this data." ``` ## Pros and Cons | 🟢 Pros | 🔴 Cons | | --- | --- | | Lowest token cost — no example overhead | Less reliable for niche or highly specific output formats | | Fastest to write and iterate on | Performance degrades on tasks outside the model's training | | Works immediately — no example curation required | Requires very precise wording to avoid ambiguity | | Best starting point before more complex techniques | Not suitable for tasks requiring domain-specific demonstrations | ## Zero-Shot Prompting Prompt Template Copy this template and replace each `{{ variable }}`. It uses Prompt Edit's variable syntax, so you can paste it into the app and fill the fields from a form every time you reuse it. ``` {{ task }} Context: {{ context }} Constraints: {{ constraints }} Output format: {{ output_format }} ``` ## Prompt Flows That Use Zero-Shot Prompting Zero-Shot Prompting is one step in this multi-framework flow. Each flow chains frameworks in sequence for a real-world outcome. - [The PoC Builder](https://www.promptedit.app/prompt-flow/the-poc-builder) · Software Development ## Frequently Asked Questions What is Zero-Shot Prompting? Zero-Shot Prompting is a technique where you give the AI model a clear instruction or question without providing any worked examples. The model relies entirely on its pre-trained knowledge and general instruction-following ability to produce an appropriate response. It is the default way most people interact with AI assistants. When is Zero-Shot Prompting the best choice? Zero-Shot is ideal for well-defined, common tasks like summarization, translation, simple Q&A, drafting emails, brainstorming, and explanation. It is also the right starting point before investing in few-shot examples — if zero-shot works, you save tokens and complexity. How can I improve Zero-Shot prompt results? The biggest lever is prompt specificity. Instead of 'summarize this', write 'Summarize the following article in 3 bullet points, each under 20 words, for a non-technical audience.' Adding constraints (length, format, audience, tone) dramatically improves zero-shot output quality without any examples. What is the difference between Zero-Shot and Few-Shot prompting? Zero-Shot provides no examples and relies on clear instructions alone. Few-Shot includes 2–8 worked input-output examples to demonstrate the exact pattern the model should follow. Few-Shot is generally more accurate for format-specific or niche tasks, but Zero-Shot is faster and cheaper. Does Zero-Shot Prompting work without instruction-tuned models? Raw base models (not instruction-tuned) respond poorly to zero-shot task instructions — they tend to continue the text rather than follow the instruction. Instruction-tuned models (GPT, Claude, Gemini) are trained to follow zero-shot instructions reliably, which is why they feel natural to use. What role does prompt clarity play in Zero-Shot performance? It is the most important factor. Ambiguous prompts lead to ambiguous outputs. The more precisely you specify the task, format, constraints, and context, the better the zero-shot result. Think of it as writing a spec for the model — every ambiguity is an invitation for the model to guess. Can Zero-Shot Prompting handle specialized or technical tasks? It can handle tasks that fall within the model's training distribution. For highly specialized domains (proprietary data, niche jargon, internal processes), zero-shot may produce generic or inaccurate results. In those cases, combine zero-shot with Role Prompting, Few-Shot examples, or Retrieval-Augmented Generation (RAG). ## Origin & Sources 1. Wei et al. (2021). [Finetuned Language Models Are Zero-Shot Learners](https://arxiv.org/abs/2109.01652) . arXiv / ICLR 2022. 2. Kojima et al. (2022). [Large Language Models are Zero-Shot Reasoners](https://arxiv.org/abs/2205.11916) . arXiv / NeurIPS 2022. ================================================================ # Framework Comparisons (5) ================================================================ ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-chain-of-draft [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Chain-of-Thought vs Chain of Draft Comparison # Chain-of-Thought vs Chain of Draft: Is Shorter Reasoning Good Enough? Chain of Draft keeps Chain-of-Thought's step-by-step structure but limits every step to a few words. The question is whether the token and latency savings are worth a possible small drop in accuracy for your task. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 **TL;DR:** [Chain of Draft](https://www.promptedit.app/prompt-framework/chain-of-draft) cut reasoning tokens by about 80% on GSM8K in its paper, with accuracy about 4 points lower than [Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought), and equal or better accuracy on other tasks. Use CoD for high-volume or latency-sensitive workloads with few-shot draft examples. Keep CoT when the reasoning must be readable or every point of accuracy matters. ## Side-by-Side Comparison | | Chain-of-Thought | Chain of Draft | | --- | --- | --- | | Instruction | "Think step by step…" | "Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most…" | | Step format | Full sentences and explanations | Numbers, equations, keywords | | GSM8K, GPT-4o | 95.4% · ~205 tokens · ~4.2 s | 91.1% · ~44 tokens · ~1.0 s | | GSM8K, Claude 3.5 Sonnet | 95.8% · ~190 tokens · ~3.1 s | 91.4% · ~40 tokens · ~1.6 s | | Zero-shot | Works well | Weak; needs draft-style examples | | Small models | Helps | Larger gap to CoT under ~3B parameters | | Readability | Easy for humans to audit | Terse; harder to review | | Introduced by | Wei et al., 2022 | Xu et al., 2025 | Benchmark numbers are from the Chain of Draft paper (Xu et al., 2025) and will differ on newer models and your own tasks. ## The Same Problem, Two Ways ### Chain-of-Thought ``` Q: Jason had 20 lollipops. He gave Denny some. Now he has 12. How many did he give Denny? A: Jason started with 20 lollipops. After giving some to Denny he has 12 left. The number he gave away is the difference between what he started with and what he has now: 20 - 12 = 8. So he gave Denny 8 lollipops. #### 8 ``` ### Chain of Draft ``` Q: Jason had 20 lollipops. He gave Denny some. Now he has 12. How many did he give Denny? A: 20 - x = 12; x = 8. #### 8 ``` ## Decision Guide - **Paying per token at scale, or users waiting on responses?** Chain of Draft, with 2–3 draft-style examples. - **Need an explanation a person will read or audit?** Chain-of-Thought. - **Accuracy is critical?** Chain-of-Thought, or Chain of Draft plus [Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency), then compare on a test set. - **Small or local model?** Chain-of-Thought. - **Reasoning model?** Skip explicit chains; state the problem clearly and ask for a concise answer. ## Frequently Asked Questions What is the difference between Chain-of-Thought and Chain of Draft? Both ask the model to reason step by step before answering. Chain-of-Thought lets each step be a full explanation; Chain of Draft limits each step to a minimal draft of about five words. The result is the same structure with far fewer tokens. How much cheaper is Chain of Draft? In the Chain of Draft paper, reasoning on the GSM8K math benchmark dropped from about 205 to 44 tokens per answer on GPT-4o and from about 190 to 40 on Claude 3.5 Sonnet, roughly an 80% reduction, with latency falling by about 60–75%. Is Chain of Draft less accurate? Sometimes slightly. On GSM8K accuracy fell from about 95% to about 91%. On date understanding and sports understanding tasks, Chain of Draft matched or beat Chain-of-Thought. Test both on your own tasks before switching. Which should I use with reasoning models? Neither needs much emphasis. Reasoning models already think internally before answering, so an explicit visible chain adds little. If you do want visible reasoning from them, Chain of Draft keeps the output short. Can I combine Chain of Draft with Self-Consistency? Yes, and it is a good fit. Self-Consistency samples several reasoning paths and takes a majority vote, which multiplies cost. Short drafts make sampling five or more paths affordable. ### Full Guides - [Chain-of-Thought →](https://www.promptedit.app/prompt-framework/chain-of-thought) Guide the model to reason through problems step by step, dramatically improving accuracy on complex tasks. - [Chain of Draft →](https://www.promptedit.app/prompt-framework/chain-of-draft) Reason step by step but keep each step to a terse draft of a few words, cutting reasoning tokens by up to 80% versus Chain-of-Thought. ### More Comparisons - [CO-STAR vs RISEN vs CRISPE →](https://www.promptedit.app/prompt-frameworks/compare/co-star-vs-risen-vs-crispe) Which structured framework fits system prompts, task instructions or creative briefs - [Chain-of-Thought vs Tree of Thoughts vs Self-Consistency →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-tree-of-thoughts) Linear reasoning, branching search or majority vote: when each pays off - [Zero-Shot vs Few-Shot Prompting →](https://www.promptedit.app/prompt-frameworks/compare/zero-shot-vs-few-shot) When instructions are enough and when examples are worth the tokens - [Context Engineering vs Prompt Engineering →](https://www.promptedit.app/prompt-frameworks/compare/context-engineering-vs-prompt-engineering) Why agents need the whole context window designed, not just the instruction ### Save as a Template Keep every framework as a reusable template with variables in Prompt Edit — free, offline and private.[Download Prompt Edit Free](https://www.promptedit.app/#download) ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-tree-of-thoughts [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Chain-of-Thought vs Tree of Thoughts vs Self-Consistency Comparison # Chain-of-Thought vs Tree of Thoughts vs Self-Consistency Three research-backed ways to make a language model reason better. They differ in one question: does the model follow one path, sample several complete paths, or search a tree of partial paths? By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 **TL;DR:** [Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) (one step-by-step path) is the cheap default for reasoning tasks. [Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) (several full paths plus a majority vote) buys reliability when there is one right answer. [Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) (branch, evaluate, backtrack) is for planning and search problems where an early wrong choice ruins the result. ## Side-by-Side Comparison | | Chain-of-Thought | Self-Consistency | Tree of Thoughts | | --- | --- | --- | --- | | Shape of reasoning | One linear path | Several independent linear paths | Tree of partial paths with evaluation | | How the answer is chosen | End of the single path | Majority vote across paths | Search: expand the best branches, backtrack from dead ends | | Best for | Math, logic, multi-step analysis | Questions with one verifiable answer | Planning, puzzles, strategy, design trade-offs | | Token cost | Low | Medium to high (N samples) | High (many branches and evaluations) | | Runs in one prompt? | Yes | Yes, or N separate calls | Simplified version only; full version needs orchestration code | | Introduced by | Wei et al., 2022 | Wang et al., 2022 | Yao et al., 2023 | | Difficulty | Beginner | Intermediate | Advanced | ## How Each Technique Works ### Chain-of-Thought You ask the model to show its intermediate steps before the final answer, either with worked examples (few-shot CoT) or a trigger such as "think step by step" (zero-shot CoT). Writing the steps out gives the model room to compute instead of guessing. ### Self-Consistency You generate several Chain-of-Thought answers with some randomness, then pick the final answer that appears most often. Different reasoning paths make different mistakes, so agreement between them is a useful signal of correctness. ### Tree of Thoughts You let the model propose several next steps, score them, and continue from the most promising ones, like a chess player considering candidate moves. When a branch looks hopeless, the search backtracks. This helps when the first plausible step is often the wrong one. ## Example Prompts ### Chain-of-Thought ``` A warehouse ships 1,240 orders on Monday and 18% more on Tuesday. On Wednesday it ships 150 fewer than Tuesday. How many orders did it ship in total? Think step by step, then give the final number on a line starting with "Answer:". ``` ### Self-Consistency ``` Solve the problem below three times using different reasoning. Label them Path 1–3 and end each with "Result:". Then report the result that appears most often and note any disagreement. [problem] ``` ### Tree of Thoughts (single-prompt version) ``` We need to cut cloud costs by 30% in two quarters without slowing releases. 1. Propose four distinct strategies. 2. Rate each for savings, risk and effort (1–10) and explain the scores. 3. Drop the two weakest. Break the remaining two into quarter-by-quarter steps. 4. If a step looks infeasible, backtrack to the next strategy. Output the final plan and why it beat the alternatives. ``` ## Decision Guide - **Everyday reasoning with a non-reasoning model?** Chain-of-Thought. - **One correct answer and errors are costly?** Self-Consistency with 3–5 samples. - **Planning or search where early choices matter?** Tree of Thoughts. - **Using a reasoning model?** Skip explicit CoT; consider Self-Consistency only for high-stakes answers. ## Frequently Asked Questions What is the difference between Chain-of-Thought and Tree of Thoughts? Chain-of-Thought asks the model to reason along one linear path, step by step, to a single answer. Tree of Thoughts has the model propose several candidate reasoning steps at each stage, evaluate them, and keep exploring only the most promising branches, backtracking when a branch fails. CoT is one pass; ToT is a search. How is Self-Consistency different from Tree of Thoughts? Self-Consistency samples several complete, independent Chain-of-Thought answers and takes a majority vote on the final result. Tree of Thoughts evaluates partial reasoning while it is being built and can abandon or expand branches. Self-Consistency improves reliability for problems with one correct answer; Tree of Thoughts helps with planning and search problems where early choices matter. Which technique uses the most tokens? Tree of Thoughts is usually the most expensive because it generates and evaluates several branches at each step, often across multiple calls. Self-Consistency costs roughly the price of one Chain-of-Thought answer multiplied by the number of samples. Plain Chain-of-Thought is the cheapest of the three. Do I still need Chain-of-Thought with reasoning models? Much less. Reasoning models from OpenAI, Anthropic and Google already think step by step before answering, so adding 'think step by step' rarely helps and can make answers longer. Self-Consistency and Tree of Thoughts can still add value on top of reasoning models for high-stakes or search-heavy problems. Can I run Tree of Thoughts in a normal chat interface? A simplified version, yes: ask the model to propose several approaches, rate each one, then expand the best and backtrack if it fails. The full technique from the original paper runs as a program that calls the model many times and keeps the search state outside the model. ### Full Guides - [Chain-of-Thought →](https://www.promptedit.app/prompt-framework/chain-of-thought) Guide the model to reason through problems step by step, dramatically improving accuracy on complex tasks. - [Tree of Thoughts →](https://www.promptedit.app/prompt-framework/tree-of-thoughts) Explore multiple distinct reasoning paths simultaneously and evaluate each branch before committing to a solution. - [Self-Consistency →](https://www.promptedit.app/prompt-framework/self-consistency) Generate multiple independent reasoning paths to the same question, then aggregate the most consistent answer. ### More Comparisons - [CO-STAR vs RISEN vs CRISPE →](https://www.promptedit.app/prompt-frameworks/compare/co-star-vs-risen-vs-crispe) Which structured framework fits system prompts, task instructions or creative briefs - [Zero-Shot vs Few-Shot Prompting →](https://www.promptedit.app/prompt-frameworks/compare/zero-shot-vs-few-shot) When instructions are enough and when examples are worth the tokens - [Chain-of-Thought vs Chain of Draft →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-chain-of-draft) Full reasoning or terse drafts: the accuracy, token and latency trade-off - [Context Engineering vs Prompt Engineering →](https://www.promptedit.app/prompt-frameworks/compare/context-engineering-vs-prompt-engineering) Why agents need the whole context window designed, not just the instruction ### Save as a Template Keep every framework as a reusable template with variables in Prompt Edit — free, offline and private.[Download Prompt Edit Free](https://www.promptedit.app/#download) ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-frameworks/compare/co-star-vs-risen-vs-crispe [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › CO-STAR vs RISEN vs CRISPE Comparison # CO-STAR vs RISEN vs CRISPE: Which Structured Prompt Framework Should You Use? All three frameworks break a prompt into named components, but they optimise for different things: CO-STAR for how the answer reads, RISEN for how the task is carried out, and CRISPE for exploring several variations. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 **TL;DR:** Use [CO-STAR](https://www.promptedit.app/prompt-framework/costar) when voice, tone and audience matter most (system prompts, brand content). Use [RISEN](https://www.promptedit.app/prompt-framework/risen-framework) when the task is a procedure with clear steps and constraints (analysis, planning, technical work). Use [CRISPE](https://www.promptedit.app/prompt-framework/crispe-framework) when you want several alternative outputs to compare (creative and marketing work). ## Side-by-Side Comparison | | CO-STAR | RISEN | CRISPE | | --- | --- | --- | --- | | Components | Context, Objective, Style, Tone, Audience, Response | Role, Instructions, Steps, End goal, Narrowing | Capacity/Role, Insight, Statement, Personality, Experiment | | Optimises for | How the output reads | How the task is executed | Range of alternatives | | Best for | System prompts, brand and audience-specific content | Analysis, planning, step-by-step technical tasks | Creative briefs, marketing copy, ideation | | Unique component | Separate Style and Tone | Narrowing (explicit exclusions and limits) | Experiment (ask for multiple versions) | | Output | One on-voice answer | One structured, complete answer | Several variations to choose from | | Token cost | Low to medium | Low to medium | Medium (multiple outputs) | | Difficulty | Intermediate | Intermediate | Intermediate | ## When to Use CO-STAR Pick CO-STAR when the same assistant or content type has to sound consistent across many inputs. Its distinguishing idea is that **Style** (structure, formality, density) and **Tone** (emotional register) are specified separately, and that the **Audience** is described precisely enough for the model to calibrate vocabulary and depth. That makes it the natural choice for system prompts, support assistants and brand content. ## When to Use RISEN Pick RISEN when success depends on the model following a process. **Steps** fix the order of work, the **End goal** defines what "done" means, and **Narrowing** lists what to exclude or which limits to respect. It is the most direct way to turn an internal briefing or standard operating procedure into a prompt. ## When to Use CRISPE Pick CRISPE when you do not yet know what the best answer looks like. The **Personality** component sets the voice, and **Experiment** asks the model for several versions, which turns a single prompt into a small A/B test. It is well suited to headlines, campaign angles and creative concepts. ## One Task, Three Frameworks The task: an announcement email for a new offline mode in a note-taking app. ### CO-STAR ``` Context: We make a note-taking app for consultants. Version 4 adds full offline mode. Objective: Get existing users to update and try offline mode this week. Style: Short paragraphs, one bullet list of what works offline, no jargon. Tone: Confident and practical, not hype. Audience: Busy consultants who work on trains and planes and have lost notes before. Response: Email under 150 words with a subject line and one call to action. ``` ### RISEN ``` Role: You are a product marketing manager for a B2B note-taking app. Instructions: Write the launch email for our new offline mode. Steps: 1) Subject line. 2) One-sentence hook about losing notes. 3) Three bullets on what works offline. 4) Update instructions. 5) Call to action. End goal: Existing users update to version 4 and open a note offline this week. Narrowing: Under 150 words. No discounts, no competitor mentions, no exclamation marks. ``` ### CRISPE ``` Capacity and Role: Act as a senior B2B copywriter. Insight: Our consultants often work without internet; version 4 adds full offline mode. Statement: Write the launch email announcing offline mode. Personality: Calm, confident, a little dry. Experiment: Give me three versions with different hooks: fear of loss, productivity, and freedom to work anywhere. ``` Notice how CO-STAR produces the most consistent voice, RISEN the most predictable structure, and CRISPE the widest range of options. ## Decision Guide - **Writing a system prompt or brand voice guide?** CO-STAR. - **Turning a process or checklist into a prompt?** RISEN. - **Need several creative options to choose from?** CRISPE. - **Need all three?** Start with RISEN for the task, add CO-STAR's Tone and Audience, and finish with CRISPE's Experiment line. ## Frequently Asked Questions Which is better for system prompts: CO-STAR, RISEN or CRISPE? CO-STAR is usually the best fit for system prompts because its separate Style, Tone and Audience components define how an assistant should communicate across many different user messages. RISEN is better when the system prompt mainly describes a procedure the assistant must follow. CRISPE fits when you want the assistant to offer several variations by default. What is the main difference between CO-STAR and RISEN? CO-STAR specifies how the output should read (style, tone, audience, response format). RISEN specifies how the task should be done (instructions, ordered steps, an end goal and narrowing constraints). CO-STAR is communication-first; RISEN is process-first. What does the E in CRISPE stand for? The E stands for Experiment: you ask the model for several alternative outputs instead of one, so you can compare and pick or combine the best. It is the component that most clearly separates CRISPE from CO-STAR and RISEN. Can I combine CO-STAR, RISEN and CRISPE? Yes. A common pattern is RISEN's Role, Instructions and Steps for the task, CO-STAR's Tone and Audience for the voice, and CRISPE's Experiment component to request three variations. Keep each component to one or two sentences so the combined prompt stays readable. Which framework is easiest for beginners? RISEN is the easiest to learn because its components map directly to how people already brief a colleague: who you are, what to do, in which steps, towards which goal, within which limits. CO-STAR takes a little more practice because Style and Tone are easy to confuse. ### Full Guides - [CO-STAR Framework →](https://www.promptedit.app/prompt-framework/costar) Context, Objective, Style, Tone, Audience, Response — six components popular for system prompt and content design. - [RISEN Framework →](https://www.promptedit.app/prompt-framework/risen-framework) Role, Instructions, Steps, End Goal, Narrowing — five parts covering who, what, how, outcome, and constraints. - [CRISPE Framework →](https://www.promptedit.app/prompt-framework/crispe-framework) Capacity/Role, Insight, Statement, Personality, Experiment — six components including tone and multiple-variation generation. ### More Comparisons - [Chain-of-Thought vs Tree of Thoughts vs Self-Consistency →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-tree-of-thoughts) Linear reasoning, branching search or majority vote: when each pays off - [Zero-Shot vs Few-Shot Prompting →](https://www.promptedit.app/prompt-frameworks/compare/zero-shot-vs-few-shot) When instructions are enough and when examples are worth the tokens - [Chain-of-Thought vs Chain of Draft →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-chain-of-draft) Full reasoning or terse drafts: the accuracy, token and latency trade-off - [Context Engineering vs Prompt Engineering →](https://www.promptedit.app/prompt-frameworks/compare/context-engineering-vs-prompt-engineering) Why agents need the whole context window designed, not just the instruction ### Save as a Template Keep every framework as a reusable template with variables in Prompt Edit — free, offline and private.[Download Prompt Edit Free](https://www.promptedit.app/#download) ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-frameworks/compare/context-engineering-vs-prompt-engineering [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Context Engineering vs Prompt Engineering Comparison # Context Engineering vs Prompt Engineering: What Changed and What Still Matters Prompt engineering is about writing good instructions. Context engineering is about everything the model sees at each step: instructions, tools, retrieved data, history and memory. For single prompts the difference is small; for agents it decides whether they work. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 **TL;DR:** Context engineering is a superset of prompt engineering. Prompt engineering optimises the instruction; context engineering optimises the whole context window across many turns, because models get less reliable as context grows ("context rot"). Use prompt frameworks to write clear instructions, and context engineering techniques (just-in-time retrieval, compaction, notes, sub-agents) to keep agents focused. ## Side-by-Side Comparison | | Prompt Engineering | Context Engineering | | --- | --- | --- | | Unit of work | One prompt or system prompt | The full context at every step of a task | | Main question | How do I phrase the instruction? | What should the model see right now, and what should it not see? | | Typical scope | Chat, one-shot generation | Agents, multi-turn assistants, RAG systems | | Key techniques | Frameworks like CO-STAR, RISEN, Few-Shot, Chain-of-Thought | Retrieval, tool design, history compaction, memory notes, sub-agents | | Main failure mode | Ambiguous or incomplete instructions | Too much, stale or irrelevant context ("context rot") | | Who uses it | Everyone who writes prompts | Teams building LLM applications and agents | ## Where the Term Came From In mid-2025 Andrej Karpathy publicly backed "context engineering" over "prompt engineering", arguing that people associate prompts with short task descriptions, while real LLM applications depend on carefully filling the context window. In September 2025 Anthropic published [Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents), describing it as curating and maintaining the optimal set of tokens during inference. ## Core Context Engineering Techniques - **System prompts at the right altitude:** specific enough to guide behaviour, general enough not to become brittle if-then rules. - **Tool design:** few, clearly scoped tools with unambiguous descriptions, so the model knows when to use which. - **Just-in-time retrieval:** keep references (file paths, IDs, links) in context and load the content only when needed, instead of stuffing everything in upfront. - **Compaction:** when a long task nears the context limit, summarise the history and continue from the summary. - **Structured note-taking:** let the agent write notes or to-do files outside the context window and read them back later. - **Sub-agents:** hand focused sub-tasks to agents with clean contexts that return short summaries. ## How Prompt Frameworks Fit In Context engineering does not make prompt frameworks obsolete. They are the building blocks of the instructions inside the context: - [CO-STAR](https://www.promptedit.app/prompt-framework/costar) or [RISEN](https://www.promptedit.app/prompt-framework/risen-framework) for writing a clear, compact system prompt. - [Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) for a one-line identity that sets the agent's expertise. - [ReAct](https://www.promptedit.app/prompt-framework/react-prompting) for the reason, act, observe loop that tool-using agents follow. - [Prompt Chaining](https://www.promptedit.app/prompt-framework/prompt-chaining) and [Least-to-Most](https://www.promptedit.app/prompt-framework/least-to-most) for splitting work into focused steps with small contexts. - [Reflexion](https://www.promptedit.app/prompt-framework/reflexion) for turning failures into short, persistent lessons instead of long transcripts. - [Few-Shot](https://www.promptedit.app/prompt-framework/few-shot) with a small set of diverse, canonical examples rather than a long list of edge cases. ## Practical Checklist - Can every sentence in the system prompt be traced to a behaviour you need? - Does each tool have a clear, non-overlapping purpose? - Is retrieved content loaded only when the current step needs it? - Do long tasks have a compaction or note-taking strategy? - Are reusable instructions saved as templates, so every run starts from the same tested context? ## Frequently Asked Questions What is context engineering? Context engineering is the practice of deciding everything that goes into a model's context window at each step: system prompt, tool definitions, retrieved documents, conversation history, memory and examples. Anthropic describes it as curating and maintaining the optimal set of tokens during inference. It matters most for agents that run over many turns. Is context engineering replacing prompt engineering? No, it contains it. Writing clear instructions (prompt engineering) is still the core of any system prompt. Context engineering adds the surrounding decisions: what to retrieve, what history to keep or summarise, which tools to expose and how to keep the context small enough to stay reliable. Who coined the term context engineering? The term spread in mid-2025, notably after Andrej Karpathy endorsed 'context engineering' over 'prompt engineering' for industrial LLM apps, describing it as the art and science of filling the context window with just the right information. Anthropic published a detailed engineering guide on it in September 2025. What is context rot? Context rot is the observation that a model's ability to recall and use information declines as the context window fills up. It is why more context is not automatically better, and why techniques like just-in-time retrieval, compaction and sub-agents exist. Which prompt frameworks are most relevant to context engineering? Agent and workflow techniques: ReAct for tool use, Prompt Chaining for splitting work across focused calls, Reflexion for persistent lessons, and Role Prompting for concise system prompts. Structured frameworks like CO-STAR or RISEN remain useful for writing the system prompt itself. ### Full Guides - [ReAct Prompting →](https://www.promptedit.app/prompt-framework/react-prompting) Interleave reasoning traces with concrete actions — Thought → Action → Observation — to solve tasks requiring tool use. - [Prompt Chaining →](https://www.promptedit.app/prompt-framework/prompt-chaining) Break complex tasks into sequential prompts where each output feeds into the next, building sophisticated pipelines. - [Reflexion →](https://www.promptedit.app/prompt-framework/reflexion) LLMs generate verbal self-reflection on their own failures and use the linguistic feedback to improve subsequent attempts. - [Role Prompting →](https://www.promptedit.app/prompt-framework/role-prompting) Assign a specific role or persona to the model to unlock domain expertise and a consistent communication style. ### More Comparisons - [CO-STAR vs RISEN vs CRISPE →](https://www.promptedit.app/prompt-frameworks/compare/co-star-vs-risen-vs-crispe) Which structured framework fits system prompts, task instructions or creative briefs - [Chain-of-Thought vs Tree of Thoughts vs Self-Consistency →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-tree-of-thoughts) Linear reasoning, branching search or majority vote: when each pays off - [Zero-Shot vs Few-Shot Prompting →](https://www.promptedit.app/prompt-frameworks/compare/zero-shot-vs-few-shot) When instructions are enough and when examples are worth the tokens - [Chain-of-Thought vs Chain of Draft →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-chain-of-draft) Full reasoning or terse drafts: the accuracy, token and latency trade-off ### Save as a Template Keep every framework as a reusable template with variables in Prompt Edit — free, offline and private.[Download Prompt Edit Free](https://www.promptedit.app/#download) ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-frameworks/compare/zero-shot-vs-few-shot [Home](https://www.promptedit.app/) ›[Prompt Frameworks](https://www.promptedit.app/prompt-frameworks) › Zero-Shot vs Few-Shot Prompting Comparison # Zero-Shot vs Few-Shot Prompting: When to Use Which Zero-shot prompting tells the model what to do. Few-shot prompting shows it. Knowing when to switch from one to the other is the most practical prompt engineering decision you make every day. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 **TL;DR:** Start with [zero-shot](https://www.promptedit.app/prompt-framework/zero-shot): a clear instruction, context and output format. Switch to [few-shot](https://www.promptedit.app/prompt-framework/few-shot) when the output format, labels or style must match exactly, or when the task is hard to describe in words. Two to five varied examples are usually enough. ## Side-by-Side Comparison | | Zero-Shot | Few-Shot | | --- | --- | --- | | What you provide | Instruction (plus context and format) | Instruction plus 2–5 worked examples | | Best for | Summaries, rewrites, translation, Q&A, drafting | Classification, extraction, strict formats, house style | | Format control | Described in words | Demonstrated by examples | | Token cost per call | Low | Higher (examples sent every time) | | Main risk | Output format drifts | Model over-copies example quirks | | Setup effort | Minimal | Curating good, varied examples | ## When Zero-Shot Is Enough Instruction-tuned models are trained to follow plain instructions, so zero-shot works for most everyday tasks. It fails mainly when you have an unusual format in mind, domain-specific labels, or a style that is easier to show than to describe. Before adding examples, try making the instruction more explicit: name the audience, the length and the exact output structure. ## When to Switch to Few-Shot - The output must match a fixed schema, label set or template exactly. - You tried a more explicit zero-shot prompt and the format still drifts. - The task depends on a house style or judgement call that is hard to put into words. - You are classifying or extracting at scale and need consistent results. ## Example Prompts ### Zero-shot ``` Classify the sentiment of this customer review as Positive, Negative or Mixed. Reply with the label only. Review: "Setup took ages, but support fixed it in ten minutes and it has worked perfectly since." ``` ### Few-shot ``` Classify the sentiment of each review as Positive, Negative or Mixed. Review: "Arrived broken and nobody answered my emails." Label: Negative Review: "Great battery life, terrible keyboard." Label: Mixed Review: "Exactly what I needed, and it shipped a day early." Label: Positive Review: "Setup took ages, but support fixed it in ten minutes and it has worked perfectly since." Label: ``` ## Tips for Better Few-Shot Examples - Vary length, wording and labels so the model learns the task, not a surface pattern. - Put the hardest or most ambiguous case among the examples. - Keep the example format identical to the format of the real input. - Save the examples as a template so every teammate uses the same set. ## Frequently Asked Questions What is the difference between zero-shot and few-shot prompting? Zero-shot prompting gives the model only an instruction. Few-shot prompting adds a handful of worked input and output examples before the real input, so the model can copy the pattern, format and level of detail shown in the examples. How many examples should a few-shot prompt have? Two to five examples cover most tasks. Add more only when the output still drifts, and make sure the examples are varied: if every example has the same length or label, the model will over-copy that feature. When is zero-shot prompting enough? When the task is common and the desired output is easy to describe in words: summaries, translations, rewrites, explanations and general questions. Modern instruction-tuned models handle these reliably without examples. Is one-shot prompting the same as few-shot? One-shot is the special case of few-shot with exactly one example. It is useful for showing a format, but a single example is easy to over-copy, so two or three varied examples are usually safer. Do few-shot examples cost more? Yes. Every example is sent with every request, so a few-shot prompt costs more tokens per call than a zero-shot prompt. For high-volume use, keep examples short or use prompt caching where your provider supports it. ### Full Guides - [Zero-Shot Prompting →](https://www.promptedit.app/prompt-framework/zero-shot) Give clear instructions without any examples, relying entirely on the model's pre-trained knowledge. - [Few-Shot Prompting →](https://www.promptedit.app/prompt-framework/few-shot) Provide a handful of input-output examples so the model learns the exact pattern or format you need. ### More Comparisons - [CO-STAR vs RISEN vs CRISPE →](https://www.promptedit.app/prompt-frameworks/compare/co-star-vs-risen-vs-crispe) Which structured framework fits system prompts, task instructions or creative briefs - [Chain-of-Thought vs Tree of Thoughts vs Self-Consistency →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-tree-of-thoughts) Linear reasoning, branching search or majority vote: when each pays off - [Chain-of-Thought vs Chain of Draft →](https://www.promptedit.app/prompt-frameworks/compare/chain-of-thought-vs-chain-of-draft) Full reasoning or terse drafts: the accuracy, token and latency trade-off - [Context Engineering vs Prompt Engineering →](https://www.promptedit.app/prompt-frameworks/compare/context-engineering-vs-prompt-engineering) Why agents need the whole context window designed, not just the instruction ### Save as a Template Keep every framework as a reusable template with variables in Prompt Edit — free, offline and private.[Download Prompt Edit Free](https://www.promptedit.app/#download) ================================================================ # Prompt Flows (15) ================================================================ ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-api-designer [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The API Designer Software Development Intermediate Prompt Flow # The API Designer Specify the full contract before the first endpoint is built, pressure-test the design from every stakeholder angle, then decompose into a sequenced implementation pipeline where nothing falls through the cracks. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[CRISP](https://www.promptedit.app/prompt-framework/crisp) Complete API contract specification 2. 2[Multi-Agent Debate](https://www.promptedit.app/prompt-framework/multi-agent-debate) Stakeholder pressure-testing of the design 3. 3[Prompt Chaining](https://www.promptedit.app/prompt-framework/prompt-chaining) Sequenced implementation pipeline ## What is The API Designer? The API Designer is a 3-step prompt flow for designing APIs that will be used, maintained, and extended in the real world — not just APIs that work in the happy path during development. It chains **CRISP** to force complete contract specification before any code is written, **Multi-Agent Debate** to pressure-test the design from the perspectives of the consumer, security reviewer, and future maintainer simultaneously, and **Prompt Chaining** to decompose the validated design into a sequenced implementation pipeline where each step's output feeds the next. The flow targets the most expensive API design failures: incomplete contracts that break consumer integrations, security assumptions that are never made explicit, and implementation gaps that emerge because the design was never sequenced. Specify first, debate second, implement in order third. ## When to Use The API Designer 🌐 ### Public-Facing APIs External APIs where breaking changes have downstream impact on integrations you don't control and can't easily patch. 🤝 ### Cross-Team Contracts Internal APIs between teams where the consumer team needs a stable contract to build against independently. 📱 ### Mobile Backend APIs APIs consumed by mobile apps where breaking changes require app store releases and can't be coordinated quickly. 🔄 ### API Versioning Designing a v2 API alongside a v1 that must remain stable — where the migration path needs to be planned before implementation. 🔐 ### Auth-Heavy APIs APIs with complex authorization models — multi-tenant, role-based, or resource-level permissions — where security must be designed in, not added later. 🧩 ### Microservice Interfaces Service-to-service APIs in a microservices architecture where contract drift causes distributed system failures. ## The Flow Algorithm 1 ### [CRISP](https://www.promptedit.app/prompt-framework/crisp) — Specify the Full Contract Use CRISP to produce a complete API specification before any implementation discussion: Clarity (every endpoint, method, path parameter, query parameter, and request body field with types and optionality), Relevance (the business operations this API enables and what each endpoint maps to), Intent (why specific design decisions were made — why POST not PUT, why this resource model), Specificity (every HTTP status code and the exact condition that triggers it, the auth mechanism and token format, the versioning strategy), Precision (error response schemas with machine-readable codes, pagination contract, rate limiting behavior). A spec that a developer who has never heard of this API could implement from alone. Produces: A complete API contract document covering all endpoints, request/response models, error schemas, auth, versioning, and design rationale — detailed enough to generate an OpenAPI spec from directly. 2 ### [Multi-Agent Debate](https://www.promptedit.app/prompt-framework/multi-agent-debate) — Pressure-Test the Design Run the API contract through a structured Multi-Agent Debate with three personas: the API Consumer (a developer integrating this API into their application — what's confusing, what's missing, what will break their use cases), the Security Reviewer (what attack surface does this expose? what are the authorization edge cases? what's missing from the auth design?), and the Future Maintainer (what will be painful to extend in 18 months? where will versioning become a problem? what naming will cause confusion as the API grows?). Each persona produces specific objections. A judge role synthesizes them into a revised contract addressing the highest-severity issues. Produces: A stress-tested, revised API contract with documented rationale for design decisions — including the objections that were considered and why they were or weren't incorporated. 3 ### [Prompt Chaining](https://www.promptedit.app/prompt-framework/prompt-chaining) — Implement in Sequence Decompose the validated API contract into an explicit implementation chain where each step's output becomes the next step's input: (1) data models and schema definitions, (2) route handler signatures and middleware stack, (3) auth implementation using the models from step 1, (4) request validation using the schemas from step 1, (5) error handling using the error contract, (6) integration tests against the contract. Each chain link is a separate, focused prompt — preventing the "implement everything at once" failure mode where late steps make incompatible assumptions about early ones. Produces: A complete, consistent API implementation where every layer was built against the contract and each step was informed by the previous one — with no auth bolted on after the fact and no error handling as an afterthought. ## Example Prompt Sequence Step 1 — CRISP Contract Specification ``` Produce a complete API contract for a user management API using CRISP: Clarity: Specify every endpoint (method, path, parameters), every request body field (name, type, required/optional), every response field, and the pagination contract for list endpoints. Relevance: The API must support: user registration, login, profile management, and admin user listing. Map each business operation to specific endpoints. Intent: Document the rationale for resource model decisions — why these resources, why these methods, why this URL structure. Specificity: List every HTTP status code used and the exact condition that triggers it. Define the JWT token format, where it goes (header/cookie), and the expiry strategy. Define the versioning strategy (/v1/ prefix or header?). Precision: Define the error response schema with a machine-readable code field. Define the rate limiting behavior (what endpoint, what limit, what happens when exceeded). Define the pagination model (cursor or offset?). ``` Step 2 — Multi-Agent Debate Design Review ``` Review the API contract below from three perspectives. Each persona should produce 3-5 specific objections with proposed changes: Persona A — API Consumer: You are a developer integrating this API into a React app. What is confusing, missing, or will break your implementation? Focus on: response field naming, error handling ergonomics, auth token refresh flow, and missing endpoints. Persona B — Security Reviewer: You are a security engineer doing a pre-launch review. What attack surface does this expose? Focus on: authorization edge cases, missing validation, token security, and privilege escalation paths. Persona C — Future Maintainer: You will be extending this API in 18 months without the original designer. What will be painful? Focus on: versioning problems, naming that won't scale, undocumented constraints, and missing extension points. Then: Judge role — synthesize the objections into a prioritized list of required changes and produce a revised contract summary. API Contract from Step 1: [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — Prompt Chaining Implementation ``` Implement the user management API using a chained sequence. Each step must use the output of the previous step. Step 3.1 — Data Models: Implement the User model, request/response DTOs, and validation schemas from the contract. Output: TypeScript interfaces and Zod schemas. Step 3.2 — Route Handlers: Using the models from 3.1, implement the route handler signatures and Express router. Output: route file with handler stubs and middleware stack. Step 3.3 — Auth Middleware: Using the User model from 3.1 and the token contract from the spec, implement JWT auth middleware. Output: middleware function. Step 3.4 — Request Validation: Using the Zod schemas from 3.1, implement validation middleware for each endpoint. Output: validation middleware. Step 3.5 — Error Handling: Using the error schema from the spec, implement the global error handler and per-route error cases. Output: error handler and error utility functions. Revised contract from Step 2: [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - Contract-first prevents the most expensive design mistakes - Multi-persona debate catches blind spots before implementation - Prompt Chaining ensures implementation consistency across layers - Produces design rationale documentation as a by-product - Works for REST, GraphQL, gRPC, or any API style ### Trade-offs - Significantly more upfront work than jumping to implementation - Debate step can produce conflicting advice that requires judgment - Prompt Chaining requires managing 5+ sequential prompts - Over-engineered for internal CRUD APIs with a single consumer ## Frequently Asked Questions What is The API Designer prompt flow? The API Designer is a 3-step prompt flow that chains CRISP, Multi-Agent Debate, and Prompt Chaining to produce API designs that survive real-world use. It specifies the full contract before writing code, pressure-tests the design through adversarial stakeholder debate, then decomposes the validated design into a sequenced implementation pipeline. Why specify the full API contract before any implementation? The most expensive API design mistakes — wrong resource models, missing error codes, broken versioning strategy — are all detectable at the contract level before a line of code is written. CRISP forces complete specification: every endpoint, every request/response model, every error schema, the auth mechanism, and the versioning strategy. Discovering these gaps in a design document costs nothing; discovering them in code review costs a full redesign. Why use Multi-Agent Debate for API design review? A self-review of an API design tends to validate your own assumptions. Multi-Agent Debate simulates the three stakeholders who will most forcefully identify design flaws: the API consumer (who has to work with the contract), the security reviewer (who sees the attack surface), and the future maintainer (who will extend it in 18 months). Each persona produces objections the original designer didn't consider, and the judge synthesizes a revised design. What does Prompt Chaining add that a single implementation prompt doesn't? A single 'implement this API' prompt produces a monolithic result where later steps make assumptions about earlier ones. Prompt Chaining explicitly sequences the implementation: data models first, then route handlers, then auth middleware, then error handling, then tests — where each step receives the previous step's output as input. This prevents the common gap where the API is designed but auth is bolted on as an afterthought. Should I use this flow for internal APIs or only public ones? Both. The contract specification and debate steps are just as valuable for internal APIs — the consumer is a different team instead of an external developer, but the cost of a broken contract is the same. Internal APIs often get less design rigor because 'we can just change it later,' which is exactly when the debate step's future maintainer persona is most valuable. How detailed should the CRISP specification be in Step 1? Specific enough that a developer who has never heard of this API could implement it from the spec alone. Every field name, type, and optionality. Every HTTP status code and what triggers it. The auth token format and where it goes. The versioning strategy and how breaking changes are handled. If you can't write it down in Step 1, you haven't designed it yet. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-brand-voice-engineer [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Brand Voice Engineer Brand Strategy Intermediate Prompt Flow # The Brand Voice Engineer Build a brand voice system that scales — extract voice identity from existing content, encode it as a reusable prompt layer, then anchor production to approved exemplars so quality never drifts. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[PIVO](https://www.promptedit.app/prompt-framework/pivo) Voice identity extraction 2. 2[DEEP](https://www.promptedit.app/prompt-framework/deep) Reusable system prompt encoding 3. 3[Few-Shot](https://www.promptedit.app/prompt-framework/few-shot) On-brand exemplar anchoring ## What is The Brand Voice Engineer? The Brand Voice Engineer is a 3-step flow for building a complete brand voice system — not just guidelines, but an executable prompt architecture that produces on-brand content at scale. It chains **PIVO** to extract and document voice identity from existing content, **DEEP** to encode that voice as a reusable system prompt layer, and **Few-Shot** to anchor production output to approved exemplars that prevent voice drift. The critical insight: voice consistency at scale requires more than a style guide. PIVO converts the style guide into a documented brief. DEEP converts the brief into a prompt. Few-Shot converts the prompt into content that actually matches approved examples. Each step produces something the previous step cannot. ## When to Use The Brand Voice Engineer 🏷️ ### Brand Launch Establishing a new brand's communication identity from scratch before producing any public-facing content. 🔄 ### Brand Refresh Documenting and encoding a refreshed voice identity when the brand's positioning or audience has evolved. 📈 ### Content Scale-Up When content production volume increases and maintaining consistent voice without individual oversight becomes necessary. 🤝 ### Agency Handoff Creating a voice brief and system prompt that external writers or agencies can use to produce on-brand content independently. 🌍 ### Multi-Channel Consistency Ensuring voice consistency across social, email, ads, website, and docs when each channel is managed separately. 🔍 ### Voice Audit Identifying and correcting voice drift in an existing content library before it affects brand perception. ## The Flow Algorithm 1 ### [PIVO](https://www.promptedit.app/prompt-framework/pivo) — Extract Voice Identity Apply PIVO to analyze the brand's existing content (or desired positioning if no content exists): Problem (what communication problem is the brand trying to solve — confusion with competitors, lack of trust, unclear differentiation), Insights (what patterns emerge from analyzing 5-10 pieces of the brand's best existing content — vocabulary choices, sentence length, what they never say), Voice (the explicit voice archetype — specific adjectives with examples, not vague descriptions like "professional"), Outcome (what a reader should feel after reading any piece of brand content). Be brutally specific in the Voice component. Produces: A Voice Identity Brief with explicit personality attributes, vocabulary preferences, and explicit "never say" examples. The brief is documentation — the executable prompt comes in Step 2. 2 ### [DEEP](https://www.promptedit.app/prompt-framework/deep) — Encode as System Prompt Feed the Voice Identity Brief into a DEEP structure to create a reusable system prompt: Direction (the brand's communication mission — one sentence), Existing Info (what the model should know about the brand, product, and audience), Expertise (the voice persona — the specific expert or archetype the brand's voice embodies), Preferred Tone (concrete tone guidance with examples: "write like this, not like this"). The output is a system prompt designed to be prepended to any content generation request. Produces: A reusable system prompt layer that encodes the brand voice. Prepend this to any content prompt to consistently activate the brand's voice without re-explaining it each time. 3 ### [Few-Shot Prompting](https://www.promptedit.app/prompt-framework/few-shot) — Anchor to Exemplars Generate (and approve) 2-3 pieces of on-brand content using the DEEP system prompt. These become your quality anchors. For all future content production, include these exemplars explicitly: "Here are 3 examples of on-brand [content type]. Write a new [content type] in the same voice." Review exemplars quarterly and update them when the voice system is refined. Produces: On-brand content with voice fidelity validated through exemplar matching rather than description-following. Voice drift is prevented at scale. ## Example Prompt Sequence Step 1 — PIVO Voice Extraction ``` Apply PIVO to extract a brand voice brief from these content samples: [PASTE 5 PIECES OF EXISTING BRAND CONTENT] Problem: What communication problem does this brand's voice solve? What would a reader feel without this voice — confused, distrustful, uninterested? Insights: Analyze the samples. What vocabulary patterns appear? What sentence lengths? What do they never say? What emotions does the content avoid? Voice: Define the voice with 5 specific adjectives. For each adjective, give one example sentence that embodies it and one that violates it. Outcome: Complete this sentence: "After reading any piece of our content, our reader should feel ___." Output: A 300-word Voice Identity Brief. ``` Step 2 — DEEP System Prompt Encoding ``` Convert the Voice Identity Brief below into a DEEP system prompt: Direction: Write a 1-sentence brand communication mission (what we communicate and why it matters to readers). Existing Info: What does the model need to know about the brand, product, and audience to write on-brand? (3-5 sentences) Expertise: What specific expert persona embodies our voice? Name the archetype specifically. Preferred Tone: Write 3 "sound like this" examples and 3 "never sound like this" examples. Output: A system prompt designed to be prepended to any content generation request. Max 200 words. Voice Identity Brief from Step 1: [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — Few-Shot Production ``` [PASTE DEEP SYSTEM PROMPT FROM STEP 2 HERE] Here are 3 examples of our on-brand LinkedIn posts: Example 1: [APPROVED EXEMPLAR] Example 2: [APPROVED EXEMPLAR] Example 3: [APPROVED EXEMPLAR] Write 3 new LinkedIn posts in exactly the same voice, tone, and structure as these examples. Topics: 1. Our new feature announcement for [FEATURE NAME] 2. A thought leadership post on [INDUSTRY TREND] 3. A customer success story for [USE CASE] ``` ## Pros and Cons ### Strengths - Produces an executable voice system, not just guidelines - DEEP system prompt is reusable across all content types - Few-Shot prevents voice drift at any scale - Works for established brands (extraction) and new brands (construction) - External writers can use the system without brand training ### Trade-offs - Requires existing content samples for PIVO extraction step - Exemplars need human approval before using as Few-Shot anchors - Voice systems need periodic auditing to catch slow drift - DEEP system prompt can grow long — monitor token budget ## Frequently Asked Questions What is The Brand Voice Engineer prompt flow? The Brand Voice Engineer chains PIVO, DEEP, and Few-Shot to build a complete brand voice system — from analysis to encoding to scaled production. PIVO extracts voice identity from existing content, DEEP encodes it as a reusable prompt layer, and Few-Shot anchors production output to exemplars to prevent voice drift. What is a 'voice identity brief' and why does PIVO produce it? A voice identity brief is a documented description of how a brand communicates — its personality traits, vocabulary preferences, tone range, and what it never sounds like. PIVO's Voice component forces this to be explicit. Without a documented brief, voice consistency depends on individual writer instinct and breaks down immediately at scale. What does DEEP encode that the PIVO brief doesn't already contain? PIVO documents the voice; DEEP translates it into prompt architecture. DEEP's Direction, Existing Info, Expertise, and Preferred Tone components turn the voice brief into a specific, structured system prompt that can be prepended to any content generation task. The brief is documentation; the DEEP output is an executable prompt. How does Few-Shot prevent voice drift? Voice drift happens when content is generated without concrete examples of what 'correct' sounds like. The DEEP system prompt describes the voice; the Few-Shot examples show it. The model matches what it sees more reliably than what it's told. Anchoring each generation to 2-3 approved exemplars prevents the gradual divergence that happens over hundreds of pieces. Can this flow work for a brand that has no existing content? Yes, with a modification. In Step 1, instead of extracting voice from existing content, use PIVO to construct the desired voice from scratch: define the brand's personality archetype, competitive positioning, and communication values. The rest of the flow works identically — you're building a new voice rather than documenting an existing one. How often should I update the voice system built by this flow? Revisit the DEEP system prompt and Few-Shot exemplars when the brand's positioning changes, when you notice consistent voice drift in generated content, or every 6-12 months as a quality audit. The PIVO voice brief should be version-controlled so changes are deliberate and documented, not accidental. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-code-architect [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Code Architect Software Development Intermediate Prompt Flow # The Code Architect Design before you code, constrain before you implement, and audit before you ship — a 3-step flow that produces architecturally sound, self-reviewed code from any feature description. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) Architecture reasoning trace 2. 2[RISEN](https://www.promptedit.app/prompt-framework/risen-framework) Constrained implementation prompt 3. 3[Reflexion](https://www.promptedit.app/prompt-framework/reflexion) Self-review & rewrite loop ## What is The Code Architect? The Code Architect is a 3-step prompt flow for software development tasks that are too complex for a single "write me this function" prompt. It chains **Chain-of-Thought** to reason through architecture before writing code, **RISEN** to structure constrained implementation prompts per component, and **Reflexion** to self-audit the result before it ships. The flow addresses the most common AI coding failure: generating code that compiles and looks plausible, but hasn't accounted for the full system — missing edge cases, incorrect assumptions about state, or architectural choices that cause problems three components later. Design before code; constrain implementation; audit the fix. ## When to Use The Code Architect 🏗️ ### Multi-Component Features New features touching multiple services, models, or layers where architectural decisions cascade through the implementation. 🔌 ### API Integrations Third-party API integrations with authentication flows, error handling, rate limiting, and retry logic to reason through. 🗄️ ### Data Pipelines ETL processes, data transformations, and batch jobs where data integrity and edge cases must be mapped before coding. 🧮 ### Complex Algorithms Sorting, graph traversal, dynamic programming, or any algorithm where correctness proof precedes implementation. 🔐 ### Auth & Security Logic Authentication flows, authorization checks, and security-sensitive code where edge case analysis is critical. ♻️ ### Refactoring Restructuring existing code where the risk of breaking behavior must be mapped before any changes are made. ## The Flow Algorithm 1 ### [Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) — Reason Through Architecture Before writing a line of code, use CoT to reason step by step through the full architecture: data models, component interfaces, state management, edge cases, failure modes, and implementation sequence. Explicitly instruct the model to flag blockers and not skip cases that seem unlikely. Produces: A detailed architecture reasoning trace with explicit trade-offs, edge cases documented, and a sequenced implementation plan where dependencies are visible. 2 ### [RISEN](https://www.promptedit.app/prompt-framework/risen-framework) — Implement Each Component For each component identified in Step 1, create a RISEN prompt: Role (senior engineer in your stack), Instructions (implement this component as spec'd), Steps (the implementation sequence), End Goal (working component with tests), Narrowing (tech stack, constraints, patterns to follow). Feed the relevant portion of the CoT architecture into each RISEN prompt. Produces: Structured, scope-bounded implementation for each component, with the architecture reasoning from Step 1 informing every decision. 3 ### [Reflexion](https://www.promptedit.app/prompt-framework/reflexion) — Self-Audit the Code After each component is generated, apply Reflexion: "Review this code critically. Does it correctly implement the spec from Step 1? What bugs, edge cases, or assumption violations exist? Does it integrate correctly with the other components? Rewrite any sections that fail this audit." Produces: Self-corrected code with an explicit audit trail — the model documents what was wrong and why the rewrite is correct, giving you a code review artifact alongside the code itself. ## Example Prompt Sequence Step 1 — Chain-of-Thought Architecture ``` Think step by step through implementing a rate limiter middleware for a Node.js Express API. Consider: 1. Data model: what state needs to be tracked per user/IP? 2. Storage: in-memory vs Redis — trade-offs for a stateless deployment? 3. Algorithm: sliding window vs token bucket — which fits a "100 requests per minute" requirement? 4. Edge cases: clock skew, distributed deployment, burst behavior 5. Failure modes: what happens if the rate limit store is unavailable? 6. Implementation sequence: what must be built first? Do not skip edge cases that seem unlikely. Flag any blockers explicitly. ``` Step 2 — RISEN Implementation ``` Role: Senior Node.js engineer familiar with Express middleware and Redis. Instructions: Implement the sliding window rate limiter middleware described in the architecture below. Steps: 1) Redis client setup, 2) sliding window algorithm, 3) Express middleware function, 4) error handling for Redis unavailability, 5) Jest unit tests. End Goal: Production-ready middleware that handles 100 req/min per IP with graceful Redis failure fallback. Narrowing: TypeScript, Express 4.x, ioredis, Jest. No external rate-limit libraries. Follow repository patterns: async/await, no callbacks. Architecture from Step 1: [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — Reflexion Audit ``` Review the rate limiter code below critically: 1. Does the sliding window implementation correctly handle the 100 req/min boundary — no off-by-one? 2. Does the Redis failure fallback actually fail open (allow requests) or fail closed (block all)? 3. Are the Jest tests testing behavior, not implementation details? 4. Are there race conditions in the Redis operations? 5. Does the TypeScript typing prevent runtime errors at the boundaries? For each issue found, explain what is wrong and provide the corrected code. [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - Architecture reasoning prevents design-time mistakes - RISEN constraints prevent scope creep per component - Reflexion audit catches bugs before they reach review - Produces an architecture document as a by-product - Works across any language or framework ### Trade-offs - Significantly more prompts than direct coding - CoT reasoning can be verbose — requires filtering - Reflexion may over-review simple, correct code - Overkill for utility functions and boilerplate ## Frequently Asked Questions What is The Code Architect prompt flow? The Code Architect is a 3-step prompt flow that chains Chain-of-Thought, RISEN, and Reflexion to produce architecturally sound, self-reviewed code. It prevents the common failure of generating code that looks correct but doesn't account for the full system. Why use Chain-of-Thought before writing any code? The most common AI coding failure is jumping straight to implementation without reasoning about architecture, edge cases, or dependencies. Chain-of-Thought forces the model to think through the full system before a single line of code is written — catching design problems before they become code problems. What does RISEN add that CoT alone doesn't provide? CoT produces a reasoning trace, not a structured implementation plan. RISEN constrains the implementation phase with explicit Role, Instructions, Steps, End Goal, and Narrowing parameters. It turns the architecture reasoning into bounded, executable prompts for each component with clear scope limits. How does Reflexion differ from just asking 'review this code'? A simple code review asks the model to find bugs. Reflexion asks the model to critically evaluate whether the fix addresses the root cause, what the fix might break elsewhere, and what tests should verify the repair. It's a post-implementation audit, not just a syntax check. Is this flow suitable for all programming tasks? The Code Architect is best suited for non-trivial features: multi-component systems, API integrations, data pipelines, and algorithms with meaningful edge cases. For simple utility functions or boilerplate, the overhead is not justified — just use direct prompting. Can I use this flow for refactoring existing code? Yes. In Step 1, feed the existing code into the CoT prompt and ask it to reason about architectural improvements, technical debt, and refactoring risks. The RISEN and Reflexion steps work identically for refactoring as for new feature development. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-debate-judge [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Debate Judge Decision-Making Advanced Prompt Flow # The Debate Judge Pressure-test contested decisions with adversarial debate, validate the verdict independently, and audit the assumptions it rests on — a 3-step flow for decisions where getting it wrong has real consequences. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Multi-Agent Debate](https://www.promptedit.app/prompt-framework/multi-agent-debate) Adversarial pressure-testing 2. 2[Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) Independent verdict validation 3. 3[FOCUS](https://www.promptedit.app/prompt-framework/focus) Assumption audit & strategy ## What is The Debate Judge? The Debate Judge is a 3-step flow for high-stakes decisions where the right answer isn't obvious and the cost of being wrong is significant. It chains **Multi-Agent Debate** to pressure-test both positions with adversarial arguments, **Self-Consistency** to validate the winning argument through independent reasoning, and **FOCUS** to map the assumptions the decision depends on. The flow addresses a fundamental weakness in single-pass AI analysis: confirmation bias. When you ask an AI "should we do X?", it tends to argue for whatever position the framing implies. Multi-Agent Debate forces steelmanning of both sides; Self-Consistency validates through independent reasoning; FOCUS ensures you know what must remain true for the recommendation to hold. ## When to Use The Debate Judge 🏗️ ### Architecture Decisions Microservices vs. monolith, build vs. buy, which database, which cloud provider — decisions that are expensive to reverse. 📈 ### Strategic Choices Market entry timing, pricing model selection, expansion decisions, and pivot-or-persist questions. 🤝 ### Partnership Decisions Vendor selection, partnership vs. build, integration strategy where each option has genuine advocates. 📋 ### Policy Design Internal policy trade-offs, process redesign options, and organizational structure decisions. 💰 ### Investment Prioritization Budget allocation decisions where opportunity costs are real and competing priorities are genuinely important. 🔬 ### Research Hypotheses Evaluating competing explanations for observed phenomena when evidence is genuinely ambiguous. ## The Flow Algorithm 1 ### [Multi-Agent Debate](https://www.promptedit.app/prompt-framework/multi-agent-debate) — Pressure-Test Both Sides Assign three explicit personas: Advocate A (makes the strongest possible case for Option 1, in good faith), Advocate B (makes the strongest possible case for Option 2, in good faith), and a Skeptical Challenger (identifies the weakest assumptions in both arguments). Run two rounds: first, each persona presents their core argument; second, each responds to the strongest point raised against them. Instruct the model to steelman — argue as if winning the debate matters. Produces: A set of steelmanned arguments and counter-arguments that surfaces objections neither you nor a single AI pass would have generated. Now you know the strongest case for each option. 2 ### [Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) — Validate the Verdict Generate three independent verdict prompts that reason from the original question and context — without reference to the debate transcript. Each should arrive at a recommendation by reasoning from first principles. Then ask: "Across these three independent analyses, which option has the most consistent support? Where do they diverge, and why?" High convergence means high confidence; divergence reveals where the decision genuinely depends on assumptions. Produces: An evidence-weighted recommendation with explicit confidence level, validated by independent reasoning paths. Divergence in the three paths is a signal, not a failure. 3 ### [FOCUS](https://www.promptedit.app/prompt-framework/focus) — Map the Assumptions Apply FOCUS to the winning option: Function (what this decision is designed to achieve), Outcome (the specific measurable result expected), Criteria (how success will be judged), Underlying Assumptions (the conditions that must remain true for this option to succeed — make these explicit, specific, and testable), Strategy (the implementation approach). The Underlying Assumptions component is the critical output: it tells you exactly what must hold for the recommendation to be correct. Produces: A decision recommendation with its assumption map — the decision-maker can now evaluate whether they believe the assumptions, rather than just evaluating the recommendation. ## Example Prompt Sequence Step 1 — Multi-Agent Debate ``` Run a two-round debate on this decision: Decision question: Should our 12-person engineering team migrate from a monolithic Rails application to microservices over the next 12 months? Round 1 — Each persona presents their core argument (200 words each): - Advocate A: Argue strongly FOR the microservices migration. Make the strongest possible case. - Advocate B: Argue strongly AGAINST the migration. Make the strongest possible case. - Skeptical Challenger: Identify the single weakest assumption in each argument. Round 2 — Each persona responds to the strongest objection raised against them (100 words each). Context: 12-person team, 6-year-old monolith, 3 domain areas, deployment twice per week, no current infrastructure team. ``` Step 2 — Self-Consistency ``` Evaluate this decision three times independently, each from first principles: Decision: Should a 12-person team migrate from a Rails monolith to microservices in 12 months? Context: [SAME CONTEXT AS STEP 1] Analysis 1: Reason from organizational capacity and team size Analysis 2: Reason from technical risk and delivery continuity Analysis 3: Reason from long-term architecture outcomes After each analysis, state a clear recommendation (migrate / do not migrate / partial migration). Then compare: Where do the three analyses agree? Where do they diverge? What does the divergence tell us about the risk of this decision? ``` Step 3 — FOCUS Assumption Audit ``` Apply the FOCUS framework to the recommended option from Step 2: Function: What is this architectural decision designed to achieve? Outcome: What specific, measurable result is expected in 12 months? Criteria: How will we know if the migration is succeeding vs. failing? Underlying Assumptions: List every assumption the recommendation depends on. For each, rate it: (a) highly likely, (b) uncertain, (c) speculative. Flag any assumption whose failure would invalidate the entire recommendation. Strategy: Given these assumptions, what is the safest implementation sequence? ``` ## Pros and Cons ### Strengths - Adversarial debate surfaces objections single-pass misses - Multi-path validation produces genuine confidence calibration - FOCUS assumption map prevents post-decision surprise - Works for any domain — technical, strategic, organizational - Produces a decision document, not just a recommendation ### Trade-offs - High token cost — 5-8 prompt interactions - Advanced technique — requires careful persona setup - Debate quality depends on model capability - Overkill for reversible, low-stakes decisions ## Frequently Asked Questions What is The Debate Judge prompt flow? The Debate Judge chains Multi-Agent Debate, Self-Consistency, and FOCUS to produce justified, high-confidence decisions on contested questions. Three AI personas debate the question from opposing positions, multiple independent verdicts validate the winning argument, and FOCUS maps the assumptions the decision depends on. What kinds of decisions is this flow best for? The Debate Judge is best for decisions where multiple viable options exist and the stakes justify the analysis effort: technology architecture choices (microservices vs. monolith), business decisions (build vs. buy vs. partner), strategic choices (market entry timing), and any question where reasonable people disagree. It's overkill for decisions with an obvious answer. How do I set up the Multi-Agent Debate personas? Assign three specific personas: Advocate A (strongest case for Option 1), Advocate B (strongest case for Option 2), and a Skeptical Challenger who questions the assumptions of both. Run each persona's argument independently, then a second round where each responds to the others' points. The personas should argue in good faith — the goal is steelmanning both sides, not winning. Why use Self-Consistency after Multi-Agent Debate? Multi-Agent Debate produces compelling arguments but doesn't produce a verdict — it's a pressure test, not a judge. Self-Consistency runs 3 independent reasoning passes from first principles (without reference to the debate) and looks for convergence. If 3 independent paths reach the same conclusion, confidence in that conclusion is high. What does FOCUS add to the decision? FOCUS makes the decision's assumptions explicit. A strong argument for Option A may rest on assumptions (market grows 20% YoY, team can hire 3 engineers, competitor doesn't react) that the decision-maker may or may not agree with. FOCUS surfaces these assumptions as visible dependencies, so the decision-maker knows exactly what must remain true for the recommendation to hold. Can this flow handle decisions with more than two options? Yes. In the Multi-Agent Debate step, assign one advocate per option (up to 4 works well), plus one skeptical challenger for all of them. Self-Consistency then evaluates which option survives multi-path scrutiny. FOCUS is applied to the winning option. More than 4 options tends to produce unfocused debate — consider narrowing to the top 2-3 before running the flow. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-error-detective [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Error Detective Debugging Intermediate Prompt Flow # The Error Detective Generate every possible cause before committing to one, stress-test each hypothesis against evidence, then audit the fix to confirm it solves the root cause — not just the symptom. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) Exhaustive hypothesis generation 2. 2[Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) Evidence-based hypothesis pruning 3. 3[Reflexion](https://www.promptedit.app/prompt-framework/reflexion) Fix validation & audit ## What is The Error Detective? The Error Detective is a 3-step debugging flow that applies scientific method to error diagnosis. It chains **Chain-of-Thought** to generate an exhaustive hypothesis space before any fix is attempted, **Tree of Thoughts** to stress-test each hypothesis against evidence and prune implausible causes, and **Reflexion** to audit the fix and confirm it addresses the root cause rather than the symptom. The flow prevents the most expensive debugging failure: premature convergence. When you ask an AI to fix a bug directly, it anchors on the first plausible explanation and generates code that may fix the visible symptom while leaving the underlying cause intact. The Error Detective forces the model through hypothesis generation, evidence evaluation, and fix validation — in that order. ## When to Use The Error Detective 🐛 ### Mysterious Production Bugs Bugs that only appear in production, intermittently, or under specific load conditions where the cause isn't immediately apparent. 🔄 ### Regression Bugs Bugs introduced by a recent change where the relationship between the change and the failure isn't obvious. 🗄️ ### Data Corruption Issues Data integrity problems where the source of corruption is unclear and multiple systems are potential culprits. ⚡ ### Performance Degradation Unexplained slowdowns where the bottleneck could be in database, application code, infrastructure, or their interaction. 🌐 ### Integration Failures Third-party API or service integration failures where the fault could be in your code, the API, or the network layer. 🔐 ### Auth & Session Bugs Authentication and authorization failures where the security implications require root cause certainty, not just symptom resolution. ## The Flow Algorithm 1 ### [Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) — Exhaust the Hypothesis Space Present the error with full context and instruct the model to reason through every possible cause systematically. Categorize hypotheses by layer: environment (OS, runtime, configuration), inputs (request payload, user state, data format), application logic (business rules, state management, conditional branches), dependencies (libraries, services, infrastructure), and timing (race conditions, timeouts, clock drift). Explicitly instruct the model not to skip causes that seem unlikely — exhaustiveness is the goal. Do not generate any fixes yet. Produces: A numbered list of causal hypotheses organized by layer, with reasoning for why each could cause the observed behavior. Nothing is ruled out yet. 2 ### [Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) — Prune to Root Cause Take the top 3-4 hypotheses from Step 1 and branch each into three sub-investigations: (a) evidence that supports this hypothesis — what about the error description, stack trace, or behavior points to this cause? (b) evidence that argues against it — what should we see if this were the cause that we don't see? (c) the specific test to confirm or rule it out — not a generic "add logging" but the exact assertion, query, or experiment. Evaluate each branch and prune hypotheses where the against-evidence is strong. Produces: A prioritized, evidence-weighted shortlist of testable hypotheses with specific diagnostic tests. The most probable root cause is now identifiable. 3 ### [Reflexion](https://www.promptedit.app/prompt-framework/reflexion) — Validate the Fix After implementing the fix, apply Reflexion: "Review this fix critically. Does it address the root cause identified in Step 2, or only the symptom? What adjacent code could this change break? Are there edge cases this fix doesn't handle? What tests must be written to verify the fix is complete and hasn't introduced regressions? If the fix only addresses a symptom, what would a root-cause fix look like?" Treat this as a mandatory audit, not an optional review. Produces: A root-cause-verified fix with an explicit audit trail, regression risk assessment, and a test checklist. The fix is defensible, not just functional. ## Example Prompt Sequence Step 1 — Chain-of-Thought Hypothesis Generation ``` Think step by step through every possible cause of this error. Do not generate fixes yet. Error: "TypeError: Cannot read properties of undefined (reading 'user')" in a Node.js Express API Observed behavior: Error occurs on POST /api/orders, but only for about 3% of requests. The remaining 97% succeed. Last change: Added Redis-based session caching 48 hours ago Stack trace: at OrderController.create (src/controllers/orders.js:47) Environment: Node 18, Express 4.18, Redis 7, deployed on AWS ECS Generate hypotheses in these categories: - Environment (configuration, infrastructure, deployment) - Inputs (request payload, user state, timing) - Application logic (race conditions, null checks, state management) - Dependencies (Redis, session middleware, request lifecycle) Do not skip hypotheses that seem unlikely. Number each one. ``` Step 2 — Tree of Thoughts Pruning ``` Take the top 4 hypotheses from the list below and apply Tree of Thoughts to each: For each hypothesis, branch into: (a) Evidence FOR: What in the error description, stack trace, or 3% failure rate points to this cause? (b) Evidence AGAINST: What should we observe if this were the cause that we don't see in the evidence? (c) Specific test: The exact diagnostic step (a query, an assertion, a log line) that confirms or rules out this hypothesis. After evaluating all 4, rank them by likelihood. Prune any with strong against-evidence. Hypotheses from Step 1: [PASTE TOP 4 HYPOTHESES HERE] ``` Step 3 — Reflexion Fix Audit ``` I implemented the following fix for the TypeError issue: [PASTE YOUR FIX HERE] Apply Reflexion: 1. Does this fix address the root cause identified in the investigation, or does it only handle the symptom? 2. What adjacent code could this change break? List specific functions or flows. 3. What edge cases does this fix not handle? Be specific. 4. Write a Jest test suite for this fix: (a) the happy path, (b) the edge case that caused the original 3% failure, (c) the regression cases from question 2. 5. If this fix is symptomatic only: describe what a root-cause fix would require. ``` ## Pros and Cons ### Strengths - Prevents premature convergence on wrong cause - Tree of Thoughts forces evidence-based hypothesis pruning - Reflexion audit prevents symptom-only fixes from shipping - Produces an investigation log as a by-product - Works for any system type — not just code ### Trade-offs - High upfront investment for simple, obvious bugs - CoT hypothesis list can be long — requires judgment to filter - Reflexion audit may be overly thorough for low-risk fixes - Does not replace running actual tests and diagnostics ## Frequently Asked Questions What is The Error Detective prompt flow? The Error Detective chains Chain-of-Thought, Tree of Thoughts, and Reflexion to diagnose and resolve complex errors systematically. CoT generates an exhaustive hypothesis space, Tree of Thoughts stress-tests each hypothesis against evidence, and Reflexion audits the fix to ensure it addresses the root cause rather than the symptom. Why not just ask the AI to debug the code directly? A direct 'fix this bug' prompt causes the model to anchor on the first plausible explanation — typically the most obvious surface symptom. The Error Detective forces exhaustive hypothesis generation before any fix is attempted, exactly as a scientist would enumerate hypotheses before testing. This prevents the 'fixed the symptom, not the cause' failure pattern. When does Tree of Thoughts add value over just ranking CoT hypotheses? Ranking hypotheses is linear — it produces a priority order but doesn't stress-test each one. Tree of Thoughts branches each hypothesis into evidence for, evidence against, and a specific test. This forces the model to argue against each hypothesis as vigorously as it argues for it, which is the only reliable way to prune false positives before investing in debugging. What does Reflexion catch that code review doesn't? Standard code review asks 'does this code look correct?'. Reflexion asks 'does this fix solve the root cause, what could it break elsewhere, and what tests should validate the repair?'. It's a post-fix audit that explicitly questions whether the fix is complete — not just syntactically correct. Can this flow be used for non-code bugs? Yes. The flow works for any system error: database query failures, infrastructure incidents, process breakdowns, data quality issues, and logical errors in business workflows. The terminology adjusts (environment/inputs/state/logic applies to any system), but the three-step reasoning structure is domain-agnostic. What context should I provide in the initial prompt? For best results, provide: the exact error message or observed behavior, the expected behavior, the last change made before the error appeared, the environment details (OS, runtime version, relevant dependencies), and any previous attempts to fix it (and why they didn't work). More context produces better hypothesis coverage in Step 1. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-legacy-integrator [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Legacy Integrator Software Development Advanced Prompt Flow # The Legacy Integrator Map the legacy system's constraints before touching it, branch every viable integration path and prune those that break existing behavior, then implement within hard scope boundaries. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[TRACE](https://www.promptedit.app/prompt-framework/trace) Legacy system constraint mapping 2. 2[Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) Multi-path integration analysis 3. 3[RISEN](https://www.promptedit.app/prompt-framework/risen-framework) Scoped implementation specification ## What is The Legacy Integrator? The Legacy Integrator is a 3-step prompt flow for the hardest common software engineering scenario: adding a feature to a codebase you didn't write, with unclear documentation, hidden coupling, and real consequences if you break existing behavior. It chains **TRACE** to map the existing system as a constrained actor before writing a line, **Tree of Thoughts** to branch and evaluate multiple integration strategies against the legacy constraints, and **RISEN** to implement the safest approach within hard scope boundaries. The flow addresses the core failure mode of legacy integration: treating the existing codebase as malleable when it isn't. Understand constraints first, explore integration paths second, narrow implementation scope third. ## When to Use The Legacy Integrator 🏚️ ### Inherited Codebases Adding features to a system you inherited — where the original developers are gone and documentation is sparse or outdated. 🔧 ### Monolith Extensions Extending a monolithic application without triggering cascading changes across the existing module graph. 🧩 ### Third-Party System Integration Integrating a new service into an existing system where the integration point is tightly coupled to business logic. 📦 ### Dependency Upgrades Upgrading a major dependency that requires changes to multiple call sites in a codebase you don't fully understand. 🔀 ### Cross-Team Contributions Contributing to another team's service or library where breaking their API contracts would block their deployments. 🗃️ ### Database Schema Changes Adding columns or tables to a legacy database schema where existing queries and migrations must remain intact. ## The Flow Algorithm 1 ### [TRACE](https://www.promptedit.app/prompt-framework/trace) — Map Legacy Constraints Before planning any change, use TRACE to reason about the existing system as a constrained actor: Task (what the system currently does), Request (what you need to add), Action (what the system expects at each integration point), Context (the hidden coupling, data contracts, and behavioral guarantees the codebase relies on), Example (a concrete trace of how data flows through the relevant code paths today). Focus the Context section on what must not change, not what exists in general. Produces: A constraint map of the legacy system: coupling points, behavioral contracts that must be preserved, risk zones where changes propagate unexpectedly, and the data flow trace your new feature must work within. 2 ### [Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) — Branch Integration Paths Feed the constraint map from Step 1 into a Tree of Thoughts prompt. Generate at least three distinct integration strategies — for example: (A) add a new module with minimal coupling, (B) extend an existing class with a new method, (C) intercept at the service boundary. For each strategy, evaluate: does it preserve the behavioral contracts from Step 1? What does it require to change? What is the regression risk? Prune approaches that violate legacy constraints before selecting the winner. Produces: A ranked evaluation of integration strategies with explicit compatibility analysis against the legacy constraints from Step 1. The output is a justified selection of the safest approach, not the most elegant one. 3 ### [RISEN](https://www.promptedit.app/prompt-framework/risen-framework) — Implement Within Hard Boundaries Implement the chosen integration strategy using RISEN, with the Narrowing parameter doing critical work: list every file, interface, and method signature that must NOT be modified. Include the behavioral contracts from Step 1 as explicit constraints. The Role should be "senior engineer who understands legacy systems and prefers additive changes over modifications." The End Goal must include characterization tests for all code paths touched. Produces: A bounded, scope-limited implementation of the new feature that adds functionality without modifying legacy contracts — plus characterization tests that document and protect the existing behavior. ## Example Prompt Sequence Step 1 — TRACE Legacy Analysis ``` Analyze the following legacy codebase using TRACE: Task: What does the current user authentication system do? Map every behavior, data contract, and assumption. Request: I need to add OAuth2 login alongside the existing username/password login — without changing how existing sessions work. Action: Identify every integration point where the new OAuth flow must interact with the existing auth system. Context: What MUST NOT change — existing session tokens, the User model schema, the /login endpoint contract, any code that reads req.user? Example: Trace the exact data flow for a current successful login: from POST /login → through middleware → to session creation → to how req.user is populated in downstream routes. [PASTE RELEVANT CODE SECTIONS HERE] ``` Step 2 — Tree of Thoughts Integration Strategies ``` Evaluate three distinct strategies for adding OAuth2 login to this legacy auth system. For each strategy: 1. Describe the approach 2. List every file and method that would need to change 3. Assess: does it preserve the behavioral contracts mapped in Step 1? (session format, User schema, req.user shape) 4. Rate the regression risk: High / Medium / Low, with reasoning 5. Identify the first thing that would break if this approach has an edge case Constraint map from Step 1: [PASTE STEP 1 OUTPUT HERE] After evaluating all three, recommend the safest approach — not the most elegant — and explain why. ``` Step 3 — RISEN Scoped Implementation ``` Role: Senior engineer who specializes in adding features to legacy systems with minimal risk. You prefer additive changes over modifications, and you treat behavioral contracts as hard constraints. Instructions: Implement the OAuth2 integration strategy selected in Step 2. Steps: 1) New OAuth route handler (additive, new file), 2) OAuth callback that creates a session in the existing format, 3) Characterization tests for all existing auth paths, 4) Integration test for the new OAuth flow. End Goal: OAuth2 login works alongside existing login. All existing auth tests pass unchanged. New characterization tests document legacy behavior. Narrowing: Do NOT modify: User model schema, session middleware, /login endpoint, req.user shape, or any existing test files. All new code goes in /auth/oauth.js and /tests/oauth.test.js only. Selected strategy from Step 2: [PASTE STEP 2 RECOMMENDATION HERE] ``` ## Pros and Cons ### Strengths - TRACE constraint mapping prevents unknown behavioral contract violations - Tree of Thoughts ensures you don't commit to the first approach - RISEN narrowing creates hard scope boundaries that prevent drift - Produces characterization tests as a by-product - Works in any language or framework ### Trade-offs - More prompts and time than a direct "add this feature" approach - Requires pasting significant legacy code context into each step - Tree of Thoughts evaluation can be verbose — requires careful reading - Overkill for well-documented, well-tested, actively maintained codebases ## Frequently Asked Questions What is The Legacy Integrator prompt flow? The Legacy Integrator is a 3-step prompt flow that chains TRACE, Tree of Thoughts, and RISEN to help you add features to codebases you didn't write without breaking existing behavior. It maps legacy constraints before planning, explores all integration paths against those constraints, then implements the safest approach within hard scope boundaries. Why use TRACE to analyze legacy code instead of just asking for an analysis? TRACE forces a structured response covering Task, Request, Action, Context, and Example — which means the model must reason about the existing system as a constrained actor with specific goals, not just summarize the code. This produces coupling maps, behavior contracts, and risk zones rather than a generic code description, which is what you need before touching legacy code. Why explore multiple integration paths before implementing? In legacy codebases, the first integration approach that comes to mind is often not the safest one — it's the one that ignores the hidden coupling your analysis revealed. Tree of Thoughts forces the model to generate and evaluate at least three distinct integration strategies, each assessed against the constraints from Step 1. You commit to the approach that best preserves existing behavior, not the one that's most obvious. What does RISEN add for legacy integration that other frameworks don't? RISEN's Narrowing parameter is critical for legacy work. It lets you explicitly constrain the implementation: no changes to files outside these boundaries, no modifications to these interfaces, preserve these exact method signatures. This prevents the common failure where the model generates a clean solution that also refactors 15 surrounding functions you weren't supposed to touch. How do I handle cases where the legacy code has no tests? In Step 1 (TRACE), include a specific request to identify all code paths and data contracts that your change must preserve, even without tests. In Step 3 (RISEN), add to the Narrowing: 'Generate characterization tests for all touched code paths before implementing the feature.' This makes the absence of tests a constraint to fill, not a gap to ignore. Is this flow suitable for modernization or migration work? For incremental modernization (adding new features while gradually improving the codebase), yes — the RISEN narrowing keeps you from accidentally doing a full refactor. For large-scale migrations, consider The Code Architect instead, which is designed for greenfield architectural work. The Legacy Integrator is specifically optimized for the 'add a feature and get out' scenario. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-market-maker [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Market Maker Marketing Copy Beginner Prompt Flow # The Market Maker Build emotionally resonant marketing copy from a product description — emotion first, conversion structure second, channel targeting third — the order that consistently outperforms single-framework copy. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[PAS](https://www.promptedit.app/prompt-framework/pas) Emotional pain-solution foundation 2. 2[AIDA](https://www.promptedit.app/prompt-framework/aida) Conversion funnel structure 3. 3[PECRA](https://www.promptedit.app/prompt-framework/pecra) Channel & audience targeting ## What is The Market Maker? The Market Maker is a 3-step marketing copy flow that chains three proven copywriting frameworks in the order that produces the most compelling results. **PAS** builds the emotional foundation — the pain, its weight, and the resolution. **AIDA** structures that emotional content into a conversion funnel. **PECRA** adapts the copy for a specific channel and audience. The critical insight is sequence: emotion must precede structure. Most AI-generated marketing copy fails not because the structure is wrong, but because it lacks emotional resonance — it describes features rather than agitating pain. By building the emotional core first, the structural frameworks have compelling material to work with. ## When to Use The Market Maker 🚀 ### Product Launches New product or feature announcements where you need to convert awareness into desire before competitors frame the narrative. 📧 ### Email Campaigns Cold outreach and nurture sequences where emotional resonance in the first line determines open-to-click rates. 🌐 ### Landing Pages Conversion-focused pages where visitors arrive with a problem and need to feel understood before they consider a solution. 📱 ### Social Ads Paid social copy where attention is scarce and the hook must stop the scroll by hitting a real, recognized pain. 🎯 ### Sales Decks Slide narratives that must build problem awareness before presenting the product as the credible resolution. 📣 ### PR & Announcements Press releases and announcement copy that must communicate value to multiple audiences simultaneously. ## The Flow Algorithm 1 ### [PAS](https://www.promptedit.app/prompt-framework/pas) — Build the Emotional Foundation Start by identifying the core pain your product solves. Use PAS to articulate it with precision: Problem (the exact situation the customer is in), Agitate (the emotional and practical weight of that problem — what it costs them, what they've tried, why it persists), Solution (how your product resolves it, specifically). Do not mention features yet — stay in pain-and-resolution territory. Produces: An emotionally resonant narrative arc that makes the reader feel understood and primes them to receive the solution with openness rather than skepticism. 2 ### [AIDA](https://www.promptedit.app/prompt-framework/aida) — Structure for Conversion Feed the PAS narrative into an AIDA prompt. Ask the model to restructure the material as a conversion sequence: Attention (a hook that stops and captures — drawn from the most acute pain in the PAS), Interest (the specifics that make this problem recognizable and the solution credible), Desire (concrete benefits, proof, transformation), Action (a single, specific CTA). The emotional material from PAS now has a conversion architecture. Produces: A conversion-structured piece with a compelling hook, credible mid-section, and clear CTA — all grounded in the emotional resonance established in Step 1. 3 ### [PECRA](https://www.promptedit.app/prompt-framework/pecra) — Target for Channel Apply PECRA to adapt the AIDA copy for your specific deployment channel. Define Purpose (what this specific piece must achieve), Expectation (the exact format: word count, structure, CTA style), Context (the channel: landing page, email, ad), Request (the specific adaptation needed), Audience (the exact reader segment). Run this step separately for each channel you need copy for — the core stays the same, the targeting varies. Produces: Channel-specific, audience-targeted copy that maintains emotional resonance and conversion structure while fitting the exact constraints of its deployment context. ## Example Prompt Sequence Step 1 — PAS Foundation ``` Apply the PAS framework to this product: Product: A time-tracking tool for freelance designers Target customer: Freelance designers who consistently undercharge for their work Problem: Identify the exact situation — what does the customer experience daily? Agitate: Amplify the pain — what does this cost them financially, emotionally, professionally? What have they tried? Why hasn't it worked? Solution: Present the product as the resolution — focus on the transformation, not the features. Write 200 words. Do not mention specific features or pricing yet. ``` Step 2 — AIDA Structure ``` Restructure the copy below as an AIDA sequence for a landing page hero section: Attention: A single headline (max 12 words) that hits the most acute pain from the copy below. No puns, no cleverness — direct and specific. Interest: 2-3 sentences that make the problem feel recognized and the solution feel credible. Desire: 3 concrete benefits (not features) with one sentence each. Action: A CTA button text (3-5 words) + one supporting line. [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — PECRA Channel Adaptation ``` Adapt the copy below for a cold email: Purpose: Get a 15-minute demo booking from a freelance designer who has never heard of this tool. Expectation: Subject line (max 8 words) + 3-paragraph email body (max 150 words total) + single CTA link. Context: Cold outreach, sent from the founder. Recipient has not opted in — we have one chance. Request: Rewrite the landing page copy for cold email format. Lead with the most relatable pain. No buzzwords. Audience: Freelance designers, 25-40, working solo or in small studios, earning $50-150k/year. [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - Emotion-first approach beats feature-listing copy - AIDA structure proven across 125+ years of direct response - PECRA step reusable across multiple channels from one core - Beginner-friendly — each framework is straightforward - Produces both strategy (PAS) and execution (AIDA+PECRA) ### Trade-offs - Requires genuine understanding of customer pain to work - PAS agitation can feel manipulative if overdone - 3 separate prompt interactions add time - Less suited for aspirational/luxury positioning ## Frequently Asked Questions What is The Market Maker prompt flow? The Market Maker chains PAS (Problem, Agitate, Solution), AIDA (Attention, Interest, Desire, Action), and PECRA (Purpose, Expectation, Context, Request, Audience) to produce high-converting marketing copy. Each step builds on the last: emotional foundation first, conversion structure second, channel targeting third. Why does PAS come before AIDA? AIDA is a structural framework — it tells you where to put content (hook, intrigue, desire, action) but not what that content should be emotionally. PAS first establishes the emotional core: the real pain, its weight, and the resolution. AIDA then structures that emotional content for conversion. Without PAS, AIDA produces structurally correct but emotionally hollow copy. What does PECRA add that AIDA doesn't cover? AIDA produces good conversion copy, but it's channel-agnostic. A landing page, a cold email, and a Twitter ad all require different length, format, and tone — even for the same product. PECRA's Audience and Context components force channel-specific adaptation, turning generic copy into deployment-ready copy. Can I skip PAS and start with AIDA? You can, but you'll typically produce copy that's structurally sound but lacks emotional resonance. PAS forces you to articulate the real pain before touching structure. The discipline of agitating the problem before presenting the solution consistently produces more compelling copy than jumping straight to the hook. What types of products work best with this flow? Any product that solves a real, felt pain — SaaS tools, services, consumer products, B2B offerings. The flow performs best when there's a genuine problem worth agitating. It's less suited to luxury or aspirational products where desire is the primary driver rather than problem resolution. How do I adapt this flow for different channels? The channel adaptation happens entirely in Step 3 (PECRA). Run Steps 1-2 once, then run Step 3 multiple times with different Context and Audience parameters: one pass for the landing page, one for the email sequence, one for the social ad. The emotional and structural core stays consistent; the channel adaptation varies. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-onboarding-writer [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Onboarding Writer Documentation Beginner Prompt Flow # The Onboarding Writer Set the documentation standard first, structure every section for how people actually learn, then lock in quality with examples — documentation that teaches users, not just informs them. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) Documentation standard & voice 2. 2[IEEI](https://www.promptedit.app/prompt-framework/ieei) Pedagogical content structure 3. 3[Few-Shot](https://www.promptedit.app/prompt-framework/few-shot) Quality anchoring via examples ## What is The Onboarding Writer? The Onboarding Writer is a 3-step documentation flow that solves two persistent problems with AI-generated documentation: content that skips context to jump straight to instructions, and quality that degrades across a long documentation set. It chains **Role Prompting** to establish the documentation standard, **IEEI** to enforce a pedagogical structure in every section, and **Few-Shot** to lock in quality across scale. The flow is built around one insight: great documentation is teaching, not information transfer. The Role Prompting and IEEI steps enforce the sequential logic of teaching — context and motivation before instruction. The Few-Shot step ensures that quality remains consistent whether you're writing the third section or the thirtieth. ## When to Use The Onboarding Writer 🚀 ### Product Onboarding First-run guides and getting started sequences where users have no prior context and every friction point causes abandonment. 🔌 ### API Documentation Developer-facing API docs where every endpoint explanation must answer "what, when, why, and what next" simultaneously. 📚 ### Knowledge Bases Help center articles and internal wikis where quality consistency across hundreds of articles is the hardest problem. 🎓 ### Tutorial Series Multi-part tutorials where each lesson must build on the last and the pedagogical structure must remain consistent. 📋 ### Standard Operating Procedures Internal SOPs where the reader is an expert in their domain but new to this specific process. 💻 ### README Files Project and library README files where developers need to go from "never heard of it" to "using it in production" in one read. ## The Flow Algorithm 1 ### [Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) — Set the Standard Define a specific documentation persona with named quality values. The persona should reference a recognized documentation standard (Stripe docs, Linear changelog, Notion's help center) and specify what that standard means: no assumed context, progressive disclosure, every instruction preceded by motivation. Include the target audience's knowledge level explicitly — "readers are experienced developers but have never used this API" sets a very different bar than "readers are technical." Produces: A documentation persona and quality standard that every subsequent section is held to. The standard is explicit, not implicit — the model can't drift from what it can't see. 2 ### [IEEI](https://www.promptedit.app/prompt-framework/ieei) — Structure for Learning Apply IEEI structure to each documentation section: Inform (what this concept/feature is — one clear sentence), Explain (why it matters and when you'd need it — the motivation before the instruction), Example (a concrete, working example — not a diagram or pseudocode, a real instance), Input (the specific action the reader takes next — what they should do now). The IEEI sequence prevents the most common documentation failure: starting with "to do X, run Y" before the reader knows what X is or why they'd want it. Produces: A pedagogically structured first section draft where context is always established before instruction. Readers understand the "why" before encountering the "how." 3 ### [Few-Shot Prompting](https://www.promptedit.app/prompt-framework/few-shot) — Lock in Quality Use the approved IEEI-structured section from Step 2 (or 2-3 sections if available) as few-shot examples in all subsequent section prompts. Provide the examples explicitly in each prompt: "Here are 2 examples of our documentation style. Write the next section in the same structure and voice." The examples enforce quality through demonstration rather than description — the model matches what it sees, not what it's told. Produces: Documentation sections that maintain consistent structure, voice, and quality across the entire set — regardless of how many sections are generated. Quality doesn't drift when it's anchored to examples. ## Example Prompt Sequence Step 1 — Role Prompting ``` You are a technical writer whose documentation standard matches Stripe's API docs: zero assumed knowledge, every concept explained before it's used, every instruction preceded by a motivation sentence, and every example is a complete, runnable code snippet (not pseudocode). Your readers: Backend developers with 3+ years of experience in their language, but who have never used this API. They are skeptical — they've been burned by bad docs before. Your documentation rule: Never write an instruction before writing the reason the reader would want to follow it. With this standard in mind, write the first section of a getting-started guide for a webhook integration API. ``` Step 2 — IEEI Structure ``` Using the documentation standard from Step 1, write the "Authentication" section for the webhook integration guide using the IEEI structure: Inform: What is the authentication method (1 sentence, no jargon)? Explain: Why does this method exist — what security problem does it solve, and why would the developer care? Example: A complete, runnable code snippet showing a correctly authenticated API request. Include the exact header format. Use Python. Input: What is the single next action the reader should take after reading this section? Strictly follow the IEEI order. Do not start with code. Start with Inform. ``` Step 3 — Few-Shot for Scale ``` Here are two examples of our documentation style for the webhook integration guide: EXAMPLE 1 — Getting Started section: [PASTE STEP 1 OUTPUT HERE] EXAMPLE 2 — Authentication section: [PASTE STEP 2 OUTPUT HERE] Using exactly the same structure, voice, and level of detail as these two examples, write the "Handling Webhook Events" section. Cover: receiving the webhook payload, verifying the signature, parsing event types, and responding with a 200 status. ``` ## Pros and Cons ### Strengths - IEEI prevents the "instructions before context" failure - Role Prompting sets an explicit, referenceable quality bar - Few-Shot scales quality across large documentation sets - Beginner-friendly — each framework is straightforward - Works for any technical domain or product type ### Trade-offs - First 2 sections (Steps 1-2) require review before using as Few-Shot anchors - Few-Shot can propagate errors if exemplars have issues - IEEI adds length — appropriate for docs, not for quick references - Role persona needs to match your actual documentation standards ## Frequently Asked Questions What is The Onboarding Writer prompt flow? The Onboarding Writer chains Role Prompting, IEEI (Inform, Explain, Example, Input), and Few-Shot Prompting to produce pedagogically sound, stylistically consistent documentation. It prevents the two most common documentation failures: telling users what to do before explaining why, and writing sections that gradually drift in quality. Why does Role Prompting come first in a documentation flow? Documentation quality is set by the standards the writer holds. By establishing a specific, named documentation persona (e.g., 'technical writer at Stripe whose docs are famous for zero assumptions and progressive disclosure'), the model applies those standards to every section. Without a persona, the model defaults to a generic 'helpful assistant' voice that lacks a consistent quality bar. What does IEEI stand for and why is it good for onboarding? IEEI stands for Inform, Explain, Example, Input. It enforces a pedagogical sequence that matches how people learn: first tell them what something is (Inform), then explain why it matters (Explain), then show a concrete example (Example), then ask them to take the next action (Input). This prevents the most common documentation skip: jumping from definition straight to instructions without building context. How do I use Few-Shot in a documentation flow? Write or generate 2-3 exemplary documentation sections using the IEEI structure and your role persona. Then use these as Few-Shot examples when prompting for the remaining sections. The model matches the quality, structure, and voice of the examples, preventing the gradual quality drift that affects long documentation sets. Can this flow be used for API documentation? Yes. In the Role Prompting step, specify an API documentation persona (e.g., 'Stripe API docs author who explains every endpoint as if the reader is smart but has never seen this API'). IEEI maps perfectly to API documentation: Inform (what this endpoint does), Explain (when and why to use it), Example (working code sample), Input (the next call the user should make). How does this flow scale to large documentation sets? The scaling power comes from Few-Shot. Once you've generated and approved 2-3 IEEI-structured sections, those become your quality anchors. Every subsequent section is prompted against those exemplars, ensuring consistent structure and voice across 10, 50, or 200 pages. Update the exemplars if quality drifts. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-poc-builder [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The PoC Builder Software Development Intermediate Prompt Flow # The PoC Builder Scope the hypothesis before writing a line, build through deliberate think-act-observe cycles, then harden the rough edges so your proof-of-concept survives the demo. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Zero-Shot](https://www.promptedit.app/prompt-framework/zero-shot) Feasibility scoping and hypothesis definition 2. 2[ReAct](https://www.promptedit.app/prompt-framework/react-prompting) Iterative think-act-observe build loop 3. 3[Prompt Refinement Loop](https://www.promptedit.app/prompt-framework/prompt-refinement-loop) Demo hardening and edge case closure ## What is The PoC Builder? The PoC Builder is a 3-step prompt flow for validating technical hypotheses quickly — without building more than you need to prove the point. It chains **Zero-Shot** to scope the hypothesis and define success criteria before touching code, **ReAct** to build iteratively through think-act-observe loops, and **Prompt Refinement Loop** to harden the rough edges before the demo. The flow targets the two most common PoC failures: scoping too broadly (building a mini-product instead of a proof) and stopping too early (a prototype that crashes under the first real question). Hypothesis first, iterative construction second, targeted polish third. ## When to Use The PoC Builder 🔬 ### Technology Evaluation Proving whether a new library, framework, or cloud service can meet your technical requirements before committing the team. 🔌 ### Third-Party API Feasibility Validating that an external API returns the data you need, in the format you need, within acceptable latency and rate limits. 🤖 ### AI/ML Model Integration Demonstrating that a model can produce useful output for your specific domain before investing in fine-tuning or infrastructure. ⚡ ### Performance Hypothesis Verifying that a proposed architecture can hit latency or throughput targets before designing the full system around it. 🏁 ### Feature Spike Exploring the feasibility of a product feature fast — enough to make a build/buy/skip decision with evidence. 🏷️ ### Vendor Comparison Building identical minimal implementations against two competing vendors so the comparison is based on real behavior, not documentation. ## The Flow Algorithm 1 ### [Zero-Shot](https://www.promptedit.app/prompt-framework/zero-shot) — Scope the Hypothesis Use Zero-Shot — no examples, no prior context — to force a clean definition of what you're actually trying to prove. Prompt the model to produce: (1) a falsifiable hypothesis statement, (2) specific success criteria with measurable thresholds, (3) the absolute minimum feature set for a demo that proves or disproves the hypothesis, and (4) what you will explicitly NOT build. The constraint of no examples prevents anchoring to existing solutions and forces first-principles scoping. Produces: A one-page PoC brief: falsifiable hypothesis, measurable success criteria, minimal viable scope, and explicit out-of-scope list. This becomes the input and exit condition for the build phase. 2 ### [ReAct](https://www.promptedit.app/prompt-framework/react-prompting) — Build Iteratively Feed the PoC brief from Step 1 into a ReAct loop: Thought (what should I build next and why) → Action (generate the code or configuration) → Observation (does this work? what did I learn?). Repeat until the success criteria are met or the hypothesis is disproved. The explicit Thought and Observation steps prevent the common trap of generating large amounts of code before discovering a fundamental problem. Build the smallest piece that teaches you the most, observe the result, iterate. Produces: A working prototype with an explicit observation trail documenting what each iteration revealed about the hypothesis. The trail is as valuable as the code — it shows your stakeholders the evidence, not just the result. 3 ### [Prompt Refinement Loop](https://www.promptedit.app/prompt-framework/prompt-refinement-loop) — Harden for Demo After the prototype proves the hypothesis, apply targeted draft → critique → revise cycles to the weakest parts. Focus the critique on demo-survival: What edge cases will an evaluator try first? What error messages will look embarrassing? What questions will the prototype fail to answer? Run 2–3 refinement cycles on specific weak spots — not a full rewrite, but targeted hardening of the parts most likely to undermine the demo's credibility. Produces: A demo-ready prototype that proves the hypothesis cleanly and doesn't collapse under predictable questions — giving stakeholders confidence in the conclusion, not just the code. ## Example Prompt Sequence Step 1 — Zero-Shot Hypothesis Scoping ``` Define the scope for a proof-of-concept with no reference to existing implementations or similar projects. Hypothesis to test: "OpenAI's Embeddings API can power a semantic search feature for our support ticket database that returns more relevant results than our current keyword search." Produce: 1. A falsifiable hypothesis statement (what specific, measurable claim are we testing?) 2. Success criteria: what exact numbers or behaviors would confirm or deny the hypothesis? 3. Minimum viable demo scope: the absolute least we need to build to get evidence 4. Explicit out-of-scope: list 5 things we will NOT build in this PoC 5. What would a "failed" PoC look like — what result proves this approach doesn't work? ``` Step 2 — ReAct Build Loop ``` Use a ReAct (Reason + Act) loop to build the semantic search PoC defined below. For each step, explicitly state: - Thought: what are you building next and why this first? - Action: the actual code or configuration - Observation: what does this output tell you about the hypothesis? PoC Brief from Step 1: [PASTE STEP 1 OUTPUT HERE] Start with the smallest piece that teaches the most: embedding a sample of 50 support tickets and running 3 test queries. Do not build the full integration until the embedding quality is confirmed. ``` Step 3 — Prompt Refinement Loop Demo Hardening ``` Apply a critique-and-revise loop to the PoC below, focused specifically on demo survival. Critique it from the perspective of a skeptical stakeholder seeing it for the first time: 1. What's the first edge case they'll try? Does it work? 2. What happens when the API is slow or returns an error? 3. Are the 3 test queries too easy — do they prove the hypothesis or just show that embeddings work on obvious cases? 4. Is the result output clear enough that a non-technical person understands what was proven? For each issue: explain the problem and provide the revised code or query. [PASTE STEP 2 PROTOTYPE HERE] ``` ## Pros and Cons ### Strengths - Prevents over-building by defining scope before coding - ReAct observation trail creates an evidence artifact for stakeholders - Refinement is targeted — only hardens what matters for the demo - Works for any technology evaluation, not just software - Produces a falsifiable hypothesis — PoC can actually disprove ideas ### Trade-offs - Step 1 scoping can feel slow when you just want to start building - ReAct loops require more prompts than a single "write this" request - Not suitable for simple function implementation or boilerplate - Refinement loop can over-polish if you don't time-box it ## Frequently Asked Questions What is The PoC Builder prompt flow? The PoC Builder is a 3-step prompt flow that chains Zero-Shot, ReAct, and Prompt Refinement Loop to help you validate a technical hypothesis with minimal code. It prevents the common failure of building a PoC that is either too broad to prove anything or too rough to survive a real demo. Why start with Zero-Shot instead of directly writing code? Most PoC failures happen before a line of code is written — the hypothesis is too vague to be falsifiable, or the scope is too large to finish in time. Zero-Shot forces you to define exactly what you're proving, what success looks like, and what the absolute minimum demo requires. This scoping step alone eliminates most PoC waste. How does ReAct differ from just asking for code? ReAct interleaves explicit reasoning steps with action steps: Thought → Action → Observation → Thought → Action → Observation. Applied to PoC development, this means the model explicitly reasons about what to build next, acts by generating code, observes the result, and iterates. The observation trail shows you where the hypothesis held and where it didn't. When should I use Prompt Refinement Loop for a PoC? Use it after the core prototype works but before the demo. Prompt Refinement Loop applies draft → critique → revise cycles to specific weak spots: error handling that crashes the demo, edge cases the evaluator will try first, or unclear output that makes the result hard to interpret. It's targeted polish, not a full rewrite. Is this flow only for software PoCs? No. The PoC Builder works for any proof-of-concept output — architecture feasibility studies, data analysis hypotheses, model fine-tuning experiments, or even non-technical proofs like pricing model validation. The chain is hypothesis-driven, not technology-specific. How do I know when the PoC has proved enough? The success criteria defined in Step 1 are your exit condition. If the Step 1 output includes a specific, measurable outcome — 'query latency under 200ms on 10k records' or 'API returns valid data for 5 test cases' — then the PoC is done when that criterion is met, regardless of how much the prototype could still be improved. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-refiner [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Refiner Content Writing Intermediate Prompt Flow # The Refiner Produce publication-ready long-form content by layering expert persona, audience structure, and multi-path quality validation — three steps that together deliver what no single prompt can. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) Domain-expert persona 2. 2[CO-STAR](https://www.promptedit.app/prompt-framework/costar) Audience/tone/style structure 3. 3[Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) 3 variations → pick strongest ## What is The Refiner? The Refiner is a 3-step prompt flow designed for long-form content that cannot afford to be mediocre. It chains three complementary techniques: **Role Prompting** anchors the model in domain expertise, **CO-STAR** structures the output for your specific audience and channel, and **Self-Consistency** validates quality by generating multiple independent versions and selecting the strongest convergent formulation. The key insight behind this flow is sequence: most writers apply structure before expertise, which produces correctly formatted but generically voiced content. The Refiner reverses this by building the expert persona first, then imposing structure, then stress-testing the result. ## When to Use The Refiner 📝 ### Blog Posts & Articles Long-form articles where expert voice and consistent quality across sections are both required. 📊 ### Executive Reports High-stakes internal documents where the audience expects domain authority and precise framing. 📄 ### White Papers Authoritative long-form content for thought leadership, where a single draft is never the final draft. 📧 ### Newsletter Issues Regular long-form newsletters that must maintain consistent voice across every issue. 🎓 ### Course Content Educational material where subject mastery must be clear and the audience level precisely calibrated. 💼 ### Proposals & Pitches Business proposals where framing, tone, and persuasive structure must all hit simultaneously. ## The Flow Algorithm 1 ### [Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) — Set the Expert Persona Begin by assigning a specific, credentialed expert persona. The more specific the role (not just "an expert" but "a senior content strategist who has written for [publication] for 15 years"), the more the model activates relevant domain knowledge. Produces: An expert-voiced first draft written from a position of domain authority, with the assumptions and vocabulary of someone deeply familiar with the subject. 2 ### [CO-STAR](https://www.promptedit.app/prompt-framework/costar) — Structure for Audience Feed the Role Prompting draft into a CO-STAR prompt that explicitly defines Context (background), Objective (what this piece must achieve), Style (e.g., narrative/analytical), Tone (e.g., authoritative/conversational), Audience (specific reader profile), and Response format. Ask the model to revise the draft to meet all six dimensions. Produces: A structured revision where the expert voice from Step 1 is now calibrated to the specific reader, channel, and purpose. Generic phrasing is replaced with audience-specific framing. 3 ### [Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) — Validate Quality Run the CO-STAR-structured prompt three times independently (starting fresh each time, without showing the model previous outputs). Then ask: "Of these three versions, which formulation is most compelling, precise, and appropriate for the audience? Synthesize the strongest elements into one final version." Produces: A quality-validated final text where every sentence has competed against alternatives. Consistent strengths survive; weak phrasings are replaced. ## Example Prompt Sequence Step 1 — Role Prompting ``` You are a senior technology journalist with 12 years of experience writing for publications like Wired and MIT Technology Review. You specialize in making complex AI topics accessible to technical-but-not-specialist readers. Write a 800-word article explaining how large language models handle context windows, why it matters for users, and what the practical limits are in current models. ``` Step 2 — CO-STAR (feed in Step 1 output) ``` Revise the article below using these parameters: Context: This will be published on a developer-focused newsletter with 20,000 subscribers. Readers are software engineers who use LLMs in production but are not ML researchers. Objective: Help readers make informed decisions about which model to use for their specific use case. Style: Analytical, grounded in concrete numbers and examples rather than abstractions. Tone: Collegial — like a knowledgeable peer explaining something, not a journalist broadcasting to an audience. Audience: Software engineers, ages 25-45, daily LLM users, skeptical of hype. Response: Keep the 800-word length. Use one concrete analogy. End with 3 practical takeaways. [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — Self-Consistency (run Step 2 three times, then) ``` I ran the previous prompt three times and got three versions of the article. Here they are: [VERSION A] [VERSION B] [VERSION C] Compare these three versions. Which formulation is most precise, engaging, and appropriate for the target audience? Synthesize the strongest introduction, clearest analogy, and most actionable takeaways from all three into one final version. ``` ## Pros and Cons ### Strengths - Consistently higher quality than single-pass prompting - Expert voice prevents generic, vague content - CO-STAR ensures audience fit, not just topic coverage - Self-Consistency catches weak sentences before publishing - Scales — works for any domain with role swap ### Trade-offs - 3x token cost on Self-Consistency step - Total flow takes 4–6 prompt interactions - Overkill for short-form or low-stakes content - Self-Consistency benefits diminish on weaker models ## Frequently Asked Questions What is The Refiner prompt flow? The Refiner is a 3-step prompt flow that chains Role Prompting, CO-STAR, and Self-Consistency to produce publication-ready long-form content. It primes the model with domain expertise, structures the output for your audience, then validates quality through multi-path convergence. When should I use The Refiner? Use The Refiner when content quality cannot be left to chance — executive reports, published blog posts, white papers, and any long-form writing where a single-pass prompt produces results that are 'good enough' but not exceptional. It's overkill for quick internal notes. Why does Role Prompting come before CO-STAR? Role Prompting establishes the knowledge and voice perspective first. If you impose CO-STAR's structure before setting a persona, the model produces structurally correct but generically voiced content. Expertise first, then structure, consistently outperforms structure first. How many variations should Self-Consistency generate? Three is the practical minimum — it surfaces divergence without excessive token cost. For high-stakes content (investor communications, published articles), five variations give more signal. The key is running each independently without showing the model its previous outputs. Can I use The Refiner for short-form content? Yes, though it's most impactful on content 500+ words. For short-form (social posts, subject lines), the overhead of three steps may not be worth it. Consider The Market Maker flow instead, which is optimized for conversion-focused short copy. What models work best with this flow? Current frontier models from OpenAI (GPT), Anthropic (Claude) and Google (Gemini) all handle this flow well. The Self-Consistency step particularly benefits from capable models — weaker models tend to produce near-identical variations, which defeats the purpose of multi-path convergence. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-research-synthesizer [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Research Synthesizer Analysis & Research Advanced Prompt Flow # The Research Synthesizer Transform a complex question into a rigorous, multi-perspective synthesis — scope first to prevent sprawl, explore all branches simultaneously, then structure findings into a formal document. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[QUEST](https://www.promptedit.app/prompt-framework/quest) Scope & boundary definition 2. 2[Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) Multi-branch exploration 3. 3[RHODES](https://www.promptedit.app/prompt-framework/rhodes) Academic synthesis structure ## What is The Research Synthesizer? The Research Synthesizer is a 3-step flow for producing rigorous, multi-perspective analysis of complex topics. It chains **QUEST** to define and bound the research scope, **Tree of Thoughts** to explore the bounded space from multiple simultaneous angles, and **RHODES** to organize the multi-branch exploration into a coherent, formal synthesis document. The flow solves two common research failures: scope sprawl (exploring everything) and single-path analysis (exploring one angle deeply while ignoring others). QUEST prevents sprawl; Tree of Thoughts enforces genuine multi-perspective exploration; RHODES provides the academic rigor that turns exploration into defensible conclusions. ## When to Use The Research Synthesizer 📊 ### Strategy Analysis Market entry decisions, build-vs-buy analysis, and competitive landscape assessments requiring multiple analytical lenses. 🏛️ ### Academic Research Literature reviews, topic surveys, and research summaries where thoroughness and intellectual rigor are required. ⚖️ ### Policy Analysis Regulatory impact assessments and policy option analyses requiring perspectives across legal, economic, and social dimensions. 💼 ### Due Diligence Investment or acquisition research requiring systematic analysis across financial, operational, market, and risk dimensions. 🔬 ### Technical Evaluation Technology selection and architecture reviews where multiple approaches must be evaluated against consistent criteria. 📰 ### Research Reports Publishable research reports and white papers requiring comprehensive coverage and structured argumentation. ## The Flow Algorithm 1 ### [QUEST](https://www.promptedit.app/prompt-framework/quest) — Define the Scope Before any exploration, use QUEST to sharply define the research boundaries: Question (the precise research question — narrow enough to be answerable), Understanding (what is already known and can be assumed as background), Expectation (what form the output should take), Scope (explicit limits — what is in and out of scope), Time (the time horizon for relevance). Reject vague questions and refine until all five dimensions are specific. Produces: A sharply bounded research brief that prevents the exploration phase from sprawling. All subsequent analysis stays within these defined limits. 2 ### [Tree of Thoughts](https://www.promptedit.app/prompt-framework/tree-of-thoughts) — Explore Multiple Branches Using the QUEST brief as the bounded space, apply Tree of Thoughts to explore the research question from 4-6 simultaneous branches: typically technical, economic, human/organizational, regulatory, competitive, and historical. For each branch, evaluate the strength of evidence before deciding whether to develop or prune it. Instruct the model to be explicit about where evidence is weak versus strong. Produces: A multi-branch analysis where each perspective has been explored and evaluated independently, with weak lines of reasoning explicitly pruned and strong ones developed into findings. 3 ### [RHODES](https://www.promptedit.app/prompt-framework/rhodes) — Structure the Synthesis Feed the Tree of Thoughts analysis into a RHODES prompt that organizes findings into formal research structure: Research (background and methodology), Hypothesis (the central claim being evaluated), Objectives (what the analysis set out to answer), Development (the multi-branch findings organized thematically), Execution (the analytical approach used), Synthesis (integrated conclusions across all branches). Ask the model to make inter-branch connections explicit. Produces: A formally structured synthesis document suitable for decision memos, research reports, or academic writing — with transparent methodology and integrated conclusions. ## Example Prompt Sequence Step 1 — QUEST Scoping ``` Apply the QUEST framework to scope this research question: Topic: Adoption of AI coding assistants in enterprise software teams Question: What are the primary barriers preventing enterprise software teams from achieving sustained productivity gains from AI coding assistants, despite initial adoption? Understanding: Assume the reader knows what GitHub Copilot, Cursor, and similar tools are. Do not explain LLMs. Expectation: A decision memo (500-800 words) for a VP of Engineering making a tool investment decision. Scope: Focus on organizational and workflow barriers, not technical limitations of the AI itself. Exclude consumer/freelance use cases. Time horizon: 2024-2026. Time: Current findings only — research older than 18 months should be flagged as potentially outdated. Confirm this scope is specific enough to be researchable, or flag what needs to be narrowed further. ``` Step 2 — Tree of Thoughts Exploration ``` Explore the research question below using Tree of Thoughts. Investigate these branches simultaneously: Branch A — Organizational: Team structure, manager resistance, review processes Branch B — Workflow: How AI tools fit (or don't) into existing development workflows Branch C — Human: Developer psychology, trust, skill anxiety, adoption behavior Branch D — Economic: ROI measurement problems, cost justification, incentive structures Branch E — Technical environment: Legacy codebases, security policies, IDE compatibility For each branch: (1) identify the strongest evidence-backed finding, (2) identify where evidence is thin, (3) evaluate if this branch is load-bearing for the central question. Prune branches with weak evidence. Develop strong branches further. Scope constraints from Step 1: [PASTE QUEST SCOPE HERE] ``` Step 3 — RHODES Synthesis ``` Organize the multi-branch analysis below into a RHODES research structure for a VP Engineering decision memo: Research: 2-sentence methodology note (how this analysis was conducted) Hypothesis: The central claim that the evidence supports or refutes Objectives: The 3 questions this memo answers Development: The 3 strongest findings from the branch analysis, with supporting evidence Execution: How the findings were evaluated and cross-validated Synthesis: 3 integrated conclusions + 3 specific recommendations for the VP Keep total length 600-800 words. Use subheadings. End with a clear recommendation. Branch analysis from Step 2: [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - QUEST prevents the most common research failure: scope sprawl - Tree of Thoughts ensures genuinely multi-perspective analysis - RHODES produces publication-quality structured output - Explicit branch pruning surfaces weak vs. strong evidence - Works for any domain — business, academic, technical, policy ### Trade-offs - High token consumption — especially Tree of Thoughts step - Requires topic familiarity to write good QUEST scope - RHODES structure may feel over-engineered for short analyses - Advanced difficulty — not suitable for simple information queries ## Frequently Asked Questions What is The Research Synthesizer prompt flow? The Research Synthesizer chains QUEST, Tree of Thoughts, and RHODES to produce a rigorous, multi-perspective analysis of any complex topic. QUEST prevents scope sprawl, Tree of Thoughts explores the bounded space from multiple angles simultaneously, and RHODES structures the exploration into a formal synthesis. Why is QUEST the first step? Research without scope limits sprawls. Without QUEST, Tree of Thoughts will explore every possible branch of a topic — producing an overwhelming amount of material with no clear boundaries. QUEST forces you to define the question precisely, the scope explicitly, and the time horizon before any exploration begins. What makes Tree of Thoughts better than just asking for a comprehensive analysis? A standard 'comprehensive analysis' prompt produces a linear essay — one line of reasoning from top to bottom. Tree of Thoughts forces parallel branching: technical, economic, regulatory, historical, and human angles explored simultaneously. Weak branches are pruned; strong ones are developed. The result is genuinely multi-perspective, not a sequential argument dressed as analysis. Is RHODES only for academic writing? No. RHODES (Research, Hypothesis, Objectives, Development, Execution, Synthesis) is inspired by academic methodology but applies to any rigorous analysis: business strategy, technical decision memos, competitive landscape reports, and policy analysis. The structure forces intellectual rigor regardless of the output format. How long does this flow typically take? The three-step flow requires 3-5 prompt interactions. Steps 1 and 2 can each produce substantial outputs (500-2000 words), and Step 3 synthesizes all of it into a final document. Budget 15-30 minutes for a thorough analysis, depending on topic complexity. Can this flow be used for market research? Absolutely. QUEST scopes the market question (e.g., 'What are the key adoption barriers for B2B AI tools in regulated industries?'). Tree of Thoughts explores it from buyer psychology, regulatory, competitive, technical, and economic branches. RHODES then structures the findings into a decision memo. It's highly effective for strategic market analysis. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-sprint-planner [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Sprint Planner Project Planning Beginner Prompt Flow # The Sprint Planner Validate the goal before planning it, reason through every task and dependency, then formalize for stakeholders — from a vague objective to a scoped, executable plan in three steps. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[SMART](https://www.promptedit.app/prompt-framework/smart) Goal definition & validation 2. 2[Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) Task & dependency reasoning 3. 3[OSCAR](https://www.promptedit.app/prompt-framework/oscar) Formal project brief structure ## What is The Sprint Planner? The Sprint Planner is a 3-step flow for turning vague project goals into fully scoped execution plans. It chains **SMART** to validate the goal before any planning begins, **Chain-of-Thought** to reason exhaustively through tasks, dependencies, and risks, and **OSCAR** to structure the output as a formal brief with explicit scope and assumptions. The flow addresses the most common planning failure: planning a goal that isn't actually a goal yet. SMART is the prerequisite — if the objective doesn't survive SMART validation, no amount of planning produces a useful plan. Once the goal is real, CoT does the hard reasoning work. OSCAR converts the reasoning into the artifact stakeholders can review and align on. ## When to Use The Sprint Planner 🏃 ### Sprint Planning Breaking a quarterly goal or epic into a sprint-sized plan with tasks, dependencies, and acceptance criteria. 🚀 ### Product Launches Cross-functional launch plans where marketing, engineering, sales, and ops tasks must be coordinated and sequenced. 🔄 ### Process Redesign Operational change initiatives where the scope must be bounded and dependencies on existing processes mapped. 📊 ### Quarterly Goals OKR breakdown — taking a high-level outcome and reasoning through the key results and initiatives needed to achieve it. 🎯 ### Project Kickoffs Project initiation where scope, constraints, and assumptions must be documented before work starts. 📝 ### Research Projects Scoping research initiatives with clear deliverables, time bounds, and explicit assumption documentation. ## The Flow Algorithm 1 ### [SMART](https://www.promptedit.app/prompt-framework/smart) — Validate the Goal Run the objective through each SMART criterion: Specific (can you state exactly what will be different when this is done?), Measurable (how will you know it's achieved — what number or artifact?), Achievable (given current resources, is this realistic?), Relevant (does this directly serve the organization's current priorities?), Time-bound (what is the exact deadline?). If any criterion fails, refine the goal before proceeding. Explicitly discard vague additions like "and improve the overall process." Produces: A precisely bounded, success-measurable objective with a deadline. The foundation everything else is planned against. 2 ### [Chain-of-Thought](https://www.promptedit.app/prompt-framework/chain-of-thought) — Reason Through Execution Apply CoT with the SMART objective as input: think step by step through every task, every dependency, every resource requirement, every risk, and every blocker needed to achieve the goal. Explicitly number tasks and show dependency arrows. Flag blockers with "BLOCKER:" prefix. Do not skip tasks that seem obvious — exhaustiveness is the point. Ask the model to reason from the deadline backwards if sequence matters. Produces: A full dependency-mapped task breakdown with explicit risk callouts. All the hidden dependencies and resource requirements that structured planning alone misses. 3 ### [OSCAR](https://www.promptedit.app/prompt-framework/oscar) — Formalize the Brief Feed the CoT task breakdown into an OSCAR structure: Objective (the SMART goal from Step 1, verbatim), Scope (what is explicitly in and out of scope — be precise, not general), Constraints (fixed limits: budget, team size, timeline, technology), Assumptions (what must remain true for this plan to work — list each and flag any that are uncertain), Results (the specific deliverables and success metrics). The Scope and Assumptions sections are the critical stakeholder alignment artifacts. Produces: A stakeholder-ready project brief that documents not just what will be done, but what won't be done and what conditions the plan depends on. ## Example Prompt Sequence Step 1 — SMART Validation ``` Apply SMART criteria to this project goal and refine it until all 5 criteria pass: Goal: "Improve the developer onboarding experience for our API" For each criterion, evaluate the current goal and rewrite to fix any failure: - Specific: What exactly will be different when this is done? - Measurable: What is the number, artifact, or observable outcome? - Achievable: Given a 2-person team and 6-week timeline, is this realistic? - Relevant: Does this directly serve current Q3 priority of reducing time-to-first-API-call? - Time-bound: What is the exact done date? Output: The revised SMART goal statement (1-2 sentences) + a brief note on what was changed and why. ``` Step 2 — Chain-of-Thought Breakdown ``` Think step by step through the execution plan for the following goal: Goal: [PASTE SMART GOAL FROM STEP 1] Team: 1 developer, 1 technical writer Timeline: 6 weeks starting Monday Constraints: No new engineering infrastructure; use existing docs platform For each task: - Number it (e.g., Task 1, Task 2) - State what it produces (a document, a code change, a decision) - Identify what it depends on (e.g., "depends on Task 3") - Flag blockers with BLOCKER: prefix Work backwards from the deadline. Do not skip tasks that seem obvious. ``` Step 3 — OSCAR Brief ``` Convert the task breakdown below into an OSCAR project brief: Objective: [SMART goal from Step 1 — verbatim] Scope: List 5 things explicitly IN scope and 3 things explicitly OUT of scope (be specific — not "no new features" but "no changes to the authentication flow") Constraints: Fixed limits from the planning context Assumptions: Every assumption the plan depends on. Flag each as: (a) high confidence, (b) uncertain, (c) unvalidated Results: The 3 specific deliverables and their acceptance criteria Keep total brief under 400 words. Use headers. Suitable for a 10-minute stakeholder review. Task breakdown from Step 2: [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - SMART catches the "bad goal" problem before planning begins - CoT surfaces hidden dependencies and blockers - OSCAR produces a stakeholder-ready brief, not just a task list - Assumptions section prevents post-launch "but we assumed…" surprises - Works for any domain — technical, marketing, research, ops ### Trade-offs - SMART validation may require 2-3 refinement rounds for vague goals - CoT reasoning can be verbose on large projects — requires editing - OSCAR brief may exceed necessary detail for small tasks - Flow doesn't update itself when plans change — rerun as needed ## Frequently Asked Questions What is The Sprint Planner prompt flow? The Sprint Planner chains SMART, Chain-of-Thought, and OSCAR to turn a vague project goal into a fully scoped, stakeholder-ready execution plan. SMART forces the goal to be real, CoT reasons through every task and dependency, and OSCAR wraps the output in a formal brief with explicit scope and assumption documentation. Why does SMART come before Chain-of-Thought? Chain-of-Thought reasons about execution in exhaustive detail — but if the goal is vague, it reasons exhaustively about the wrong thing. SMART is the goal validation checkpoint: if you can't make the goal Specific, Measurable, Achievable, Relevant, and Time-bound, you're not ready to plan yet. SMART forces goal clarity before planning work begins. What does Chain-of-Thought add to project planning? Structured frameworks like OSCAR can organize a plan, but they can't discover tasks you didn't think to include. Chain-of-Thought does the exhaustive reasoning work: what tasks exist, what depends on what, what could go wrong, what resources are needed. It surfaces the hidden dependencies and risks that structured planning alone misses. What does OSCAR contribute that CoT doesn't? CoT produces a reasoning trace — a detailed but informal breakdown. OSCAR structures it into a formal brief with explicit Objective, Scope, Constraints, Assumptions, and Results sections. The Scope and Assumptions components are especially important for stakeholder communication: they document what is and isn't included, and what must remain true for the plan to work. Can this flow be used for non-technical projects? Absolutely. The flow is domain-agnostic — SMART, CoT, and OSCAR work equally well for marketing campaigns, product launches, research projects, organizational change initiatives, and any goal-oriented work. The output format adjusts to the domain; the reasoning structure is universal. How should I handle a goal that fails SMART validation? If your goal fails SMART, don't skip to planning — fix the goal first. A goal that isn't Specific probably has multiple competing definitions of success. A goal that isn't Measurable will never produce a clear done state. A goal that isn't Time-bound will expand to fill all available time. Fix these before using CoT — you'll save hours of planning work on the wrong objective. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-story-architect [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Story Architect Product Ownership Beginner Prompt Flow # The Story Architect Turn a feature idea into a Jira-ready user story with complete, testable acceptance criteria — a 3-step flow that gives every story the scenario context and restriction clarity that prevents dev interpretation drift. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) Senior PO writing standard 2. 2[COAST](https://www.promptedit.app/prompt-framework/coast) Scenario-based story structure 3. 3[CLEAR](https://www.promptedit.app/prompt-framework/clear) Logic + restrictions = AC ## What is The Story Architect? The Story Architect is a 3-step prompt flow for product owners who need to turn a feature idea into a development-ready user story — with the scenario context and acceptance criteria that prevent dev teams from building the wrong thing. It chains **Role Prompting** to set a senior PO writing standard, **COAST** to structure the story within its full scenario context, and **CLEAR** to generate the logic, expectations, and restrictions that become testable acceptance criteria. The flow addresses the two most common user story failures: stories that lack scenario context (so devs make assumptions that diverge from intent) and acceptance criteria that are too vague to be testable ("the feature should work as expected"). COAST prevents the first by forcing scenario grounding. CLEAR prevents the second by requiring explicit logic, expected behaviour, and explicit restrictions. ## When to Use The Story Architect 🆕 ### New Feature Stories Turning a product idea or stakeholder request into a well-scoped, scenario-grounded story ready for sprint planning. ♻️ ### Refining Existing Stories Improving vague backlog items that lack scenario context or have insufficient acceptance criteria before they enter a sprint. 🔀 ### Epic Decomposition Breaking a large epic into individual stories by running COAST multiple times with different scenario values for each user path. 🐛 ### Bug Reports as Stories Structuring a bug report as a formal story with reproduction scenario, expected behaviour, and explicit fix restrictions. 🏗️ ### Technical Debt Stories Writing tech debt stories with a business-justified Logic component that explains why prioritisation is warranted. 🤝 ### Stakeholder Alignment Using the COAST output as a structured format for stakeholder review before acceptance criteria are finalised. ## The Flow Algorithm 1 ### [Role Prompting](https://www.promptedit.app/prompt-framework/role-prompting) — Set the PO Standard Assign a specific senior PO persona with named quality values before writing any story content. The persona should specify the writing standard you expect: stories written in standard "As a… I want… so that…" format, acceptance criteria written as testable Given/When/Then conditions, explicit edge cases, and scope restrictions on every story. Reference a benchmark if useful — "the standard of stories that never come back from dev with clarification questions." Produces: A writing persona that applies a consistent, high-quality standard to every story generated in this session — preventing the vague, one-liner stories that cause sprint friction. 2 ### [COAST](https://www.promptedit.app/prompt-framework/coast) — Structure the Story Apply COAST to structure the feature as a fully contextualised user story: **Context** (product background, user type, current state — what is true before this story), **Objective** (what the user is trying to achieve — the "so that" clause), **Actions** (what the user does to achieve it — the "I want to" clause, broken into discrete steps), **Scenario** (the specific situation that triggered this story — the concrete use case, not the abstract capability), **Task** (the development work in scope — what engineering will build). Ask the model to output a formatted story with these components explicit. Produces: A user story grounded in a real scenario, with the user's objective, actions, and development scope all explicit — no room for dev interpretation of intent. 3 ### [CLEAR](https://www.promptedit.app/prompt-framework/clear) — Write the Acceptance Criteria Feed the COAST story into a CLEAR prompt to generate complete acceptance criteria: **Context** (re-ground in the scenario from Step 2), **Logic** (the reasoning behind each criterion — why this specific behaviour is required), **Expectations** (the specific, testable conditions that must be true — Given/When/Then per criterion), **Action** (what the system must do for each expectation to pass), **Restrictions** (what must NOT happen, what is explicitly out of scope for this story). The Restrictions component is critical — it defines the story's boundary and prevents scope creep during development. Produces: Complete, testable acceptance criteria with explicit restrictions. Every criterion has a reason (Logic), a test (Expectations), and a boundary (Restrictions) — the full definition of done. ## Example Prompt Sequence Step 1 — Role Prompting ``` You are a senior product owner with 10 years of experience writing user stories for agile software teams. Your stories are known for never coming back from dev with clarification questions. Your standards: - Every story uses the format: "As a [user type], I want [action], so that [outcome]." - Acceptance criteria are written as testable Given/When/Then conditions — never vague statements like "should work correctly." - Every story includes at least one edge case and one explicit out-of-scope restriction. - Scenario context is always specific — a concrete situation, not an abstract capability description. With these standards, help me write a user story for the following feature. ``` Step 2 — COAST Story Structure ``` Using the COAST framework, write a user story for this feature: Feature idea: Allow users to save a prompt template as a favourite so they can find it quickly. Product: A macOS app for managing reusable AI prompt templates. User: A daily AI user who has 30+ templates saved and struggles to find the ones they use most. Structure the story using: Context: Who is the user, what is their current situation, and what friction exists today? Objective: What outcome does the user want — the "so that" clause? Actions: What specific steps does the user take — the "I want to" clause? Scenario: The specific moment that triggered this need (e.g., "The user opens the app to quickly reuse their daily standup template but has to scroll through 30 templates to find it.") Task: What does engineering build — scoped to this story only? Output: A formatted user story with each COAST component labelled, followed by the one-line story sentence. ``` Step 3 — CLEAR Acceptance Criteria ``` Using the CLEAR framework, write complete acceptance criteria for the user story below. For each acceptance criterion: Context: Re-state the relevant part of the scenario that this criterion addresses. Logic: One sentence explaining why this specific behaviour is required. Expectations: Written as Given / When / Then — specific and testable, no vague language. Action: What the system does when the expectation is met. Restrictions: What must NOT happen, or what is explicitly out of scope for this story. Include criteria for: the happy path, at least one edge case (e.g., favouriting a template that was already favourited), and one explicit out-of-scope restriction. User story from Step 2: [PASTE STEP 2 OUTPUT HERE] ``` ## Pros and Cons ### Strengths - Scenario grounding prevents dev interpretation drift - CLEAR Restrictions component reduces scope creep - Given/When/Then format makes AC immediately testable by QA - Beginner-friendly — each step is straightforward - Scales: run COAST multiple times per epic to decompose stories ### Trade-offs - More verbose than a one-liner story — appropriate for complex features, not trivial tasks - COAST scenario requires a concrete use case — needs domain knowledge to write well - Step 3 AC should be reviewed before adding to Jira — not a fully automated output - Overkill for small bug fixes or copy change tickets ## Frequently Asked Questions What is The Story Architect prompt flow? The Story Architect chains Role Prompting, COAST, and CLEAR to turn a feature idea into a Jira-ready user story with complete acceptance criteria. Role Prompting sets the quality standard, COAST structures the story within its full scenario context, and CLEAR adds the logic and restrictions that become testable acceptance criteria. Why use Role Prompting first instead of jumping straight to COAST? Without a persona, the model writes user stories at a generic 'helpful assistant' standard — which typically produces structurally correct but uselessly vague acceptance criteria like 'the feature should work correctly'. A senior PO persona applies the specific standards you actually need: precise Given/When/Then logic, explicit edge cases, and scope restrictions that prevent dev interpretation drift. How does COAST map to a user story? COAST's five components align naturally with user story structure: Context (the product and user background), Objective (what the user is trying to achieve — the 'so that' clause), Actions (what the user does — the 'I want to' clause), Scenario (the specific usage situation that prompted this story), Task (the development task in scope). Together they produce a richly contextualised story rather than a one-liner. What does CLEAR add that COAST doesn't already cover? COAST produces the story and its scenario context. CLEAR adds the acceptance criteria layer: Context (re-grounds the AC in the story scenario), Logic (the reasoning behind each criterion — why this matters), Expectations (specific, testable conditions that must be true), Action (what the system must do), Restrictions (what must NOT happen, what is out of scope). The Restrictions component is particularly valuable — it prevents devs from building beyond scope. Can this flow generate multiple user stories from a single feature? Yes. Run Step 2 (COAST) multiple times with different Scenario values to generate stories for different user paths through the same feature. Each scenario produces a distinct story with its own context. Then run Step 3 (CLEAR) for each story. This is especially useful for features with multiple user types or happy/sad path distinctions. How do I handle epics versus individual stories with this flow? For an epic, run Step 1 once and Step 2 with a high-level Objective. The COAST output will naturally reveal multiple distinct Actions — each one becomes a candidate individual story. Then run Step 3 (CLEAR) on each individual story. The flow scales from epic decomposition down to single-story refinement. Can I adapt this flow for bug reports or technical debt stories? Yes. For bugs: in COAST, the Scenario is the reproduction steps, the Objective is the expected behaviour, and the Actions are the buggy steps. CLEAR's Restrictions then specify what the fix must not change. For technical debt: the Scenario is the current pain, the Objective is the improved state, and CLEAR's Logic explains the business justification for prioritising the work. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-test-strategist [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The Test Strategist Software Development Intermediate Prompt Flow # The Test Strategist Define the risk surface before writing a single test, generate multiple independent coverage strategies and converge on the most robust, then anchor output to your codebase's real test style. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[CRISPE](https://www.promptedit.app/prompt-framework/crispe-framework) Risk surface analysis as senior test engineer 2. 2[Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) Multi-path coverage strategy validation 3. 3[Few-Shot](https://www.promptedit.app/prompt-framework/few-shot) Style-anchored test generation ## What is The Test Strategist? The Test Strategist is a 3-step prompt flow for building test suites that actually catch bugs — not just test suites that pass CI and hit coverage thresholds. It chains **CRISPE** to frame the model as an experienced test engineer reasoning over the real risk surface of your code, **Self-Consistency** to generate multiple independent coverage strategies and find consensus on what matters most, and **Few-Shot** to anchor the final test output to your codebase's real style and patterns. The flow addresses the most common testing anti-pattern: generating tests bottom-up from the code, which produces tests that cover what exists rather than what could go wrong. Risk surface first, consensus strategy second, style-anchored generation third. ## When to Use The Test Strategist 🐛 ### Critical Business Logic Testing payment processing, authentication flows, data transformations, or any code where a bug has direct business impact. 🔄 ### Refactoring Coverage Adding tests before a refactor to ensure behavior is preserved — where you need confidence, not just coverage percentage. 🧩 ### Complex Integration Points Testing code at the boundary between two systems — where assumptions on both sides can be wrong simultaneously. 📊 ### Data Processing Pipelines Validating ETL steps, aggregation logic, and data transformations where edge cases in input data cause silent failures. 🔐 ### Security-Sensitive Code Testing authorization checks, input sanitization, and rate limiting where a missed edge case is a vulnerability. 🆕 ### New Team Onboarding When a new team takes over a module and needs to understand the risk surface before making changes. ## The Flow Algorithm 1 ### [CRISPE](https://www.promptedit.app/prompt-framework/crispe-framework) — Map the Risk Surface Use CRISPE to frame the analysis: Capacity (senior test engineer with experience in this domain — payment processing, auth systems, etc.), Role (your task is to identify every failure mode, not to generate tests), Insight (here is the code and its business context), Statement (produce a ranked risk map), Personality (be specific and skeptical — assume the implementation has bugs), Experiment (identify the 3 highest-risk areas and what test types would catch failures there). The Insight section should include the actual code plus business context that explains what a bug would cost. Produces: A ranked risk map: the top failure modes by likelihood and impact, the specific edge cases and boundary conditions within each, and the test types (unit, integration, property-based, etc.) most likely to catch each class of failure. 2 ### [Self-Consistency](https://www.promptedit.app/prompt-framework/self-consistency) — Validate Coverage Priorities Generate three independent coverage strategies from the same risk map using Self-Consistency: ask the model to approach the coverage question three times with different reasoning paths (e.g., from the perspective of a developer who has been paged at 3am, a security auditor, and a QA engineer doing exploratory testing). Look for consensus: which risk areas do all three strategies agree need tests? Which areas only one strategy flags? High-consensus areas get tests first; single-strategy flags get investigated before investing test time. Produces: A validated coverage strategy with confidence levels — high confidence on consensus items, flagged-for-investigation on single-strategy items. This prevents both under-testing critical paths and over-testing low-risk code. 3 ### [Few-Shot](https://www.promptedit.app/prompt-framework/few-shot) — Generate Style-Anchored Tests Provide 3 existing tests from your codebase as Few-Shot examples — ideally one simple unit test, one test with mocking, and one async or integration test. Then ask the model to generate tests for the high-consensus risk areas identified in Step 2, using the same patterns, assertion style, naming conventions, and setup/teardown structure as the examples. The Few-Shot anchoring ensures the output integrates without a style review cycle. Produces: Test cases targeting the validated highest-risk areas, written in the codebase's actual style — ready to commit alongside the code with minimal review friction. ## Example Prompt Sequence Step 1 — CRISPE Risk Surface Analysis ``` Capacity: You are a senior test engineer with 10+ years of experience writing tests for financial transaction systems. You have been paged at 3am because a bug in code like this caused a production incident. Role: Your task is to produce a ranked risk map for this code — NOT to write tests yet. Identify failure modes, edge cases, and boundary conditions. Insight: This is a discount code validation function for an e-commerce checkout. A bug here either gives users unauthorized discounts (revenue loss) or incorrectly blocks valid codes (lost sales). Statement: Produce: (1) top 5 failure modes by likelihood × impact, (2) the specific inputs or states that trigger each, (3) what test type would catch each. Personality: Be specific and skeptical. Assume there are bugs. Do not flag obvious happy-path cases. Experiment: For each failure mode, write a one-line hypothesis: "If [condition], then [failure]." [PASTE FUNCTION CODE HERE] ``` Step 2 — Self-Consistency Coverage Validation ``` Using the risk map below, generate three independent test coverage strategies. Approach the problem from three different perspectives: Perspective A: A developer who was paged at 3am when this code broke in production — what tests would have caught it? Perspective B: A security auditor looking for discount abuse vectors — what inputs would they try? Perspective C: A QA engineer doing exploratory testing on the checkout flow — what user behavior would they test? For each perspective, list the top 5 test cases in priority order. Then: identify which test cases appear in at least 2 of the 3 perspectives (high confidence) vs. only 1 (investigate further). Risk map from Step 1: [PASTE STEP 1 OUTPUT HERE] ``` Step 3 — Few-Shot Style-Anchored Generation ``` Here are 3 examples of tests from our codebase. Study the style: assertion library, naming convention, describe/it structure, mock setup pattern, and error message format. Example 1: [PASTE EXISTING TEST] Example 2: [PASTE EXISTING TEST] Example 3: [PASTE EXISTING TEST] Now write tests for the following high-confidence risk areas (from Step 2 consensus): 1. [RISK AREA 1] 2. [RISK AREA 2] 3. [RISK AREA 3] Match the style of the examples exactly. Do not introduce new assertion libraries or patterns not present in the examples. ``` ## Pros and Cons ### Strengths - Risk-driven approach catches bugs that coverage metrics miss - Self-Consistency reduces blind spots in coverage strategy - Few-Shot style anchoring eliminates review friction on generated tests - Works for any language, framework, or test runner - Produces a risk map artifact useful beyond testing ### Trade-offs - More setup than "generate tests for this function" - Requires existing test examples for optimal style anchoring - Self-Consistency step requires reviewing 3 independent outputs - Not optimized for simple utility functions with obvious test cases ## Frequently Asked Questions What is The Test Strategist prompt flow? The Test Strategist is a 3-step prompt flow that chains CRISPE, Self-Consistency, and Few-Shot to build test suites that catch real bugs rather than just hitting coverage metrics. It reasons about risk before writing tests, validates coverage priorities through multiple independent strategies, and anchors test output to the codebase's actual style using examples. Why start with risk analysis instead of just generating tests? Most AI-generated tests cover the happy path and obvious inputs — which is exactly where bugs aren't. CRISPE's Capacity and Insight parameters force the model to reason as an experienced test engineer who has seen the specific failure modes of this type of code. The output is a ranked risk map, not a list of test cases, so every test you write is justified by a specific risk. What does Self-Consistency add to test strategy? A single coverage strategy reflects one reasoning path, which may have blind spots. Self-Consistency generates three independent strategies from the same risk map and looks for consensus: if all three strategies agree that the authentication edge cases are highest priority, that consensus is high-confidence. If only one strategy flags a risk area, that's a signal to investigate further before investing tests there. Why use Few-Shot for the actual test generation? Tests that don't match the codebase's existing style create friction — they use different assertion libraries, naming conventions, or setup patterns. Few-Shot locks the generated tests to 3 existing tests from the real test suite, so the output looks like a human wrote it and integrates without style review cycles. The style anchoring is as important as the content. How many existing test examples should I provide in Step 3? Three examples is the recommended minimum — one for a simple unit test, one for a test with mocking, and one for an integration or async test if applicable. The examples should represent the actual patterns in use, not the cleanest tests in the suite. You want the model to match what's there, not produce idealized tests that conflict with the codebase's conventions. Can this flow work for a codebase with no existing tests? Yes, with a modification. In Step 3, instead of providing existing test examples, provide the testing framework documentation and 2-3 examples from the framework's own docs as your Few-Shot examples. Clearly state in the prompt that these are canonical style examples, not codebase examples. The output will match the framework's conventions, which is a reasonable starting point for a greenfield test suite. ---------------------------------------------------------------- Source: https://www.promptedit.app/prompt-flow/the-ux-strategist [Home](https://www.promptedit.app/) ›[Prompt Flows](https://www.promptedit.app/prompt-flows) › The UX Strategist Product Design Intermediate Prompt Flow # The UX Strategist Map constraints before ideating, break first-instinct design thinking with systematic creativity, then convert the best concept into an implementation-ready spec — in three steps. By [Sebastian Messingfeld](https://www.linkedin.com/in/messingfeld/), Staff Engineer · Updated September 23, 2026 1. 1[GROW](https://www.promptedit.app/prompt-framework/grow) Constraint & goal mapping 2. 2[SCAMPER](https://www.promptedit.app/prompt-framework/scamper) 7 creative design lenses 3. 3[SEED](https://www.promptedit.app/prompt-framework/seed) Deliverable specification ## What is The UX Strategist? The UX Strategist is a 3-step design flow that prevents two common design failures: building ideas disconnected from constraints, and anchoring on the first plausible concept. It chains **GROW** to map the design space within real constraints, **SCAMPER** to systematically break first-instinct thinking and generate unexpected alternatives, and **SEED** to convert the strongest concept into a deliverable-focused specification. The critical sequence insight: constraints must precede creativity. SCAMPER without GROW produces clever ideas that can't be built. SCAMPER after GROW generates creative alternatives within achievable bounds. SEED then converts ideas into commissions — artifacts engineering can actually work from. ## When to Use The UX Strategist ✨ ### Feature Redesigns Existing features that aren't meeting user needs where first-instinct solutions are too incremental to make a difference. 🆕 ### New Feature Design Brand-new features where the design space needs systematic exploration before committing to an approach. 🔀 ### User Flow Optimization Multi-step flows with friction where incremental improvements have failed and a structural rethink is needed. 📱 ### Mobile-First Design Mobile experiences where constraints are tightest and first-instinct desktop-adapted designs consistently underperform. ♿ ### Accessibility Redesign Adding accessibility where it was retrofitted, requiring creative solutions that serve all users without degrading the primary experience. 🚀 ### Onboarding Design First-run experiences where the gap between "user arrives" and "user succeeds" must be systematically reduced. ## The Flow Algorithm 1 ### [GROW](https://www.promptedit.app/prompt-framework/grow) — Map the Problem Space Apply GROW to the design problem: Goal (the user experience you're trying to achieve — framed as what the user can do/feel, not what you'll build), Reality (the current friction — what is actually happening today, with data if available), Options (2-3 design directions that are conceptually distinct — don't evaluate yet, just identify), Will (constraint audit — what can actually be built given team size, tech stack, timeline, and technical debt). The Will step is non-negotiable — remove any option that fails the constraint audit. Produces: A problem-solution space map with constraint-qualified design directions. Ideation begins from here — within achievable bounds. 2 ### [SCAMPER](https://www.promptedit.app/prompt-framework/scamper) — Break First-Instinct Thinking Apply all seven SCAMPER lenses to the current design or primary Option from Step 1. For each lens, produce one concrete design alternative (not a vague idea — a specific change to a specific UI element or interaction). Rate each alternative: viable within constraints (yes/no) and potential impact (low/medium/high). Select the 1-2 most promising alternatives for development in Step 3. Some lenses will produce weak ideas — prune them explicitly rather than keeping everything. Produces: A rated set of design alternatives, including at least one unexpected concept that wouldn't have emerged from linear design thinking. The best concept is selected for specification. 3 ### [SEED](https://www.promptedit.app/prompt-framework/seed) — Specify the Deliverable Apply SEED to the winning concept from Step 2: Situation (the user context and current state), End Goal (the specific measurable outcome — not "better UX" but "reduce time-to-first-action from 4 steps to 2"), Examples (reference designs or interaction patterns from other products), Deliverables (the exact artifacts needed: screen flows, component list, interaction states, edge case handling). The Deliverables section is the handoff artifact — specific enough that a designer can execute it without a follow-up meeting. Produces: An implementation-ready design specification with measurable success criteria, reference examples, and explicitly scoped deliverables. Ready for design tool execution or engineering handoff. ## Example Prompt Sequence Step 1 — GROW Problem Mapping ``` Apply GROW to this UX design problem: Product: A B2B SaaS project management tool Problem: New users abandon the product within the first 3 days because they can't create their first project successfully. Goal: What does successful onboarding feel like for the user? Frame as user experience, not features. Reality: Describe the current friction — what specifically causes abandonment at which step? (Infer from the problem if data isn't provided.) Options: Identify 3 conceptually distinct design directions for reducing abandonment. Will: For a 2-person design team with a 3-week timeline, which options are buildable? Remove any that fail this constraint. ``` Step 2 — SCAMPER Creative Exploration ``` Apply SCAMPER to the primary onboarding flow for the project management tool described below. For each lens, produce one specific design alternative: Substitute: Replace one element of the current flow with something from a different interaction paradigm Combine: Merge two currently separate steps into one Adapt: Borrow a pattern from a consumer app (games, social media, e-commerce) Modify: Change the scale, speed, or sequence of something Put to another use: Repurpose an existing feature in a new onboarding context Eliminate: Remove a step that isn't essential for first-time success Rearrange: Reorder the sequence for better cognitive flow For each: rate it (viable within 3-week constraint: yes/no) and (impact potential: low/medium/high). Select the 2 most promising alternatives. Current flow context from Step 1: [PASTE GROW OUTPUT HERE] ``` Step 3 — SEED Specification ``` Create a design specification using SEED for the winning concept below: Situation: New B2B user, first login, no prior context on the product. Has agreed to try it but is skeptical. On desktop browser. End Goal: User completes their first project creation with at least one task added within 5 minutes of first login. Measured by event tracking. Examples: Reference Notion's first-run template picker, Linear's empty state guidance, and Figma's template gallery as positive patterns to draw from. Deliverables: (1) Screen flow diagram (screens + transitions), (2) Component list with interaction states, (3) Empty state and error state specs, (4) Success measurement event list. Winning concept from Step 2: [PASTE SELECTED SCAMPER CONCEPT HERE] ``` ## Pros and Cons ### Strengths - GROW prevents ideating outside buildable constraints - SCAMPER systematically breaks first-instinct anchoring - SEED produces engineering-ready handoff artifacts - Measurable success criteria built into SEED output - Works for any product type or platform ### Trade-offs - Some SCAMPER lenses produce weak ideas — requires judgment to prune - SEED spec quality depends on how well constraints are defined - Does not produce visual mockups — requires design tool follow-up - 3 separate prompt interactions add time versus direct ideation ## Frequently Asked Questions What is The UX Strategist prompt flow? The UX Strategist chains GROW, SCAMPER, and SEED to produce design specifications grounded in real constraints. GROW maps the design problem within its constraints, SCAMPER generates creative alternatives that break first-instinct thinking, and SEED converts the best concept into an implementable deliverable. Why use a coaching framework (GROW) for UX design? GROW's four components — Goal (desired experience), Reality (current friction), Options (design paths), Will (what can actually be built) — are directly applicable to UX problem-framing. The Will component specifically forces constraint acknowledgment before ideation, preventing SCAMPER from generating designs that look great but can't be built. How should I apply SCAMPER to UX design? Apply each of SCAMPER's seven lenses to the current design or user flow: Substitute (replace a UI element or interaction pattern with something else), Combine (merge two separate features), Adapt (borrow a pattern from another domain), Modify (change scale, speed, or frequency), Put to another use (repurpose existing UI), Eliminate (remove an element entirely), Rearrange (change the order or layout). Some lenses will produce weak ideas — that's expected. The goal is one surprising, viable concept. What does SEED add that a design brief doesn't? SEED's four components force the design brief to be deliverable-focused rather than vision-focused. Situation (the context), End Goal (the outcome measurement), Examples (reference designs or interaction patterns), Deliverables (the specific artifacts required) — the Deliverables component is critical. It turns an idea into a commission: something engineering can actually build. Can this flow be used for mobile design? Yes. The flow is medium-agnostic. In the GROW step, specify mobile constraints (screen size, thumb zones, network variability). In SCAMPER, apply the lenses to mobile-specific interaction patterns. In SEED, specify mobile deliverables (screen flows, touch targets, component specs). How far does this flow go — does it produce actual mockups? The flow produces a design specification document — the brief that informs mockup creation, not the mockups themselves. For UI implementation, hand the SEED output to a designer or use it as input for a design tool. The flow's value is in producing a well-reasoned, constraint-grounded brief rather than in generating visual artifacts.