CREATE Framework
Six components that build prompts like a creative brief. CREATE gives AI a persona to inhabit, examples to learn from, and room for iterative refinement — treating prompt engineering as a process, not a one-shot attempt.
Introduced: CREATE was developed by Dave Birss, an AI consultant and author, in 2023. It stands out among community frameworks by including “Examples” as a core component — acknowledging the power of few-shot learning. The “Adjustments” element also encourages iterative refinement, treating prompting as a process rather than a one-shot effort. Each letter maps to one building block: Character (persona), Request (task), Examples (demonstrations), Adjustments (refinements), Type of Output (format), and Extras (additional constraints).
Modern LLM Status: CREATE remains a practical and accessible framework for users who want a structured approach to prompt construction. Its inclusion of examples as a first-class component aligns with research showing that few-shot demonstrations improve output quality significantly. The “Adjustments” step is especially relevant today: modern LLMs respond well to iterative refinement within a conversation, and CREATE formalizes that practice. Whether you use Claude, GPT-4, or Gemini, the six CREATE dimensions provide a reliable checklist for building effective prompts.
Give AI a Role, a Roadmap, and Room to Improve
Most prompts tell the AI what to do but forget to tell it who to be, what good looks like, or how to refine its approach. The result is a technically adequate but generic response that misses the mark for your specific needs. CREATE addresses this gap by treating every prompt as a six-part creative brief.
CREATE builds prompts in layers. First, you assign a Character — a persona that shapes the AI’s perspective and expertise. Then you define the Request (the core task), provide Examples (demonstrations of the desired output), specify Adjustments (refinements and constraints to iterate on), declare the Type of Output (format and structure), and add Extras (guardrails, edge cases, or additional rules). Each layer adds precision.
Think of it like briefing a freelance specialist. You would not just say “write something” — you would explain who they are writing as, show them samples of what you want, and tell them how to refine their first draft. CREATE formalizes that same natural process for AI interactions.
Most frameworks stop at describing the task and format. CREATE goes further by embedding few-shot examples directly into the prompt structure — showing the AI what “good” looks like rather than just describing it. The Adjustments layer then invites iteration: refine, constrain, and redirect without starting over. This makes CREATE especially effective for tasks where the first output is close but needs targeted refinement.
The CREATE Process
Six components that build prompts through persona, demonstration, and refinement
Character — Assign a Persona
Define who the AI should be for this task. A Character gives the model a perspective, domain expertise, and communication style to draw from. This is more than role-playing — it anchors the AI’s vocabulary, assumptions, and problem-solving approach in a specific professional identity.
“You are a veteran high school science teacher who excels at explaining complex concepts using everyday analogies.”
Request — State the Task
Clearly articulate what you need the AI to produce. The Request should be specific enough to act on but not so prescriptive that it constrains creative solutions. Focus on the outcome you need rather than dictating every step of how to get there.
“Explain the greenhouse effect to a class of 14-year-olds who have never studied climate science before.”
Examples — Show What Good Looks Like
Provide one or more demonstrations of the output quality, format, or style you expect. Examples leverage the power of few-shot learning — showing the AI a concrete pattern to follow rather than relying on abstract descriptions alone. This is CREATE’s signature differentiator from other structured frameworks.
“Here is how I explain photosynthesis: ‘Imagine a plant running a tiny solar-powered kitchen. Sunlight is the electricity, water and CO2 are the ingredients, and glucose is the meal it cooks for energy.’ Use this analogy-first approach.”
Adjustments — Refine and Constrain
Specify refinements, constraints, or modifications that shape the output toward your exact needs. Adjustments acknowledge that great prompts are iterative — the first output is a draft that you refine by adding constraints, narrowing focus, or redirecting emphasis. This step can also be used after seeing the initial response to course-correct.
“Avoid technical jargon. Use at least two real-world analogies. Keep the explanation under 200 words. Do not mention specific political policies.”
Type of Output — Define the Format
Specify the exact format, structure, and presentation of the response. Type of Output eliminates ambiguity about whether you want a paragraph, a bulleted list, a table, a script, or a slide deck outline. Being explicit about format saves revision cycles and ensures the output is immediately usable.
“Write it as a classroom script with a hook question, two analogy-based explanations, and a quick comprehension check question at the end.”
Extras — Add Guardrails and Edge Cases
Include any additional constraints, edge cases, special instructions, or guardrails that do not fit neatly into the other five components. Extras is the catch-all for anything that would refine the output further — ethical considerations, things to avoid, audience-specific sensitivities, or cross-references to other materials.
“Include a note reminding students that AI-generated explanations should be verified against their textbook. Avoid any politically charged framing of climate topics.”
See the Difference
Why six structured components produce better results than a single instruction
Casual Prompt
Write a product description for a standing desk.
Introducing our new standing desk. This adjustable desk allows you to work while standing or sitting. It features a smooth height adjustment mechanism and a spacious work surface. Perfect for improving your posture and productivity. Available in multiple colors.
CREATE Prompt
C: You are a copywriter at a premium ergonomic furniture brand known for witty, benefit-driven product pages.
R: Write a product description for our new motorized standing desk, the “Elevate Pro.”
E: Here is our style: “The Lumbar King chair does not just support your back — it makes your old office chair file for early retirement.”
A: Emphasize the whisper-quiet motor and cable management system. Do not mention competitors by name.
T: 150-word product page blurb with a headline, 3 benefit bullets, and a closing CTA.
E: Include a note that claims about health benefits should be verified with a healthcare professional. Remember to verify all AI-generated marketing copy before publishing.
Rise Above the 9-to-5 Slump
The Elevate Pro does not just adjust — it transforms your workday without making a sound. Our whisper-quiet motor moves you from sitting to standing in under 4 seconds, so seamlessly your coffee will not even ripple.
• Silent transition — 40dB motor that will not interrupt your calls
• Hidden cable channels that turn desk spaghetti into clean lines
• Memory presets for your perfect sit and stand heights
Your back has been filing complaints for years. Time to promote it. [Shop Elevate Pro]
Note: Consult a healthcare professional regarding posture-related health claims.
Practice Responsible AI
Always verify AI-generated content before use. AI systems can produce confident but incorrect responses. When using AI professionally, transparent disclosure is both best practice and increasingly a legal requirement.
48 US states now require AI transparency in key areas. Critical thinking remains your strongest tool against misinformation.
CREATE in Action
See all six components working together across different scenarios
Character: You are a senior backend engineer who writes clear, opinionated technical blog posts. You favor practical advice over theoretical deep-dives.
Request: Write a blog post explaining why our team migrated from REST to GraphQL for our mobile API.
Examples: Here is our blog voice: “We did not switch to GraphQL because it was trendy. We switched because our mobile app was making 14 API calls to render a single screen, and our users on 3G connections were not impressed.”
Adjustments: Include one code snippet showing a before/after comparison. Acknowledge trade-offs honestly — do not make GraphQL sound perfect. Mention the learning curve for the team.
Type of Output: Blog post, 600–800 words. Title, 4 sections with headers, one code block, and a “Would we do it again?” conclusion.
Extras: Do not mention specific client names. Include a note that readers should evaluate GraphQL against their own architecture before migrating. Always verify technical claims against current documentation.
Why We Broke Up with REST (and Mostly Do Not Regret It)
The Problem
Our mobile app was drowning in network requests. A single user profile screen required 14 separate REST calls — user data, preferences, activity feed, notifications, and a handful of nested resources. On fast Wi-Fi, nobody noticed. On cellular connections in emerging markets, users were staring at loading spinners for 8+ seconds...
The Migration
We did not rip out REST overnight. We ran both systems in parallel for three months, migrating one screen at a time...
Note: Evaluate GraphQL against your own architecture and team capabilities before committing to a migration. Every codebase has different constraints.
Character: You are an HR director at a mid-size tech company who values constructive, growth-oriented feedback over bureaucratic checkbox reviews.
Request: Create a performance review template that managers can use for quarterly one-on-ones with their direct reports.
Examples: Instead of “Rate communication skills 1–5,” use prompts like: “Describe a specific situation where this person’s communication made a measurable impact on a project outcome.”
Adjustments: Include sections for both manager assessment and employee self-assessment. Add space for career development goals. Avoid any language that could create legal liability.
Type of Output: Fillable template with 5 sections, each containing 2–3 guided prompts. Include instructions for managers at the top.
Extras: Note that all performance documentation should be reviewed by HR before formal delivery. Remind managers to verify any data points referenced in reviews. AI-generated templates should be adapted to your organization’s specific policies.
Quarterly Growth Review — Manager Guide
Instructions: Complete your sections before the meeting. Share with your direct report 24 hours in advance so they can prepare their self-assessment portions. This is a conversation starter, not a verdict.
Section 1: Impact & Contributions
Manager: Describe one project where this person’s work directly moved a team or company metric. What was the before and after?
Self-Assessment: What accomplishment from this quarter are you most proud of, and why?...
Note: Have HR review all completed reviews before formal delivery. Adapt this template to your organization’s specific policies and legal requirements.
Character: You are a product researcher specializing in qualitative user interviews for B2B SaaS products. You prioritize open-ended discovery over confirmation bias.
Request: Design a 30-minute customer interview guide to understand why enterprise clients are churning after the first 90 days.
Examples: Good question: “Walk me through a typical day when you use our product. Where does it fit into your workflow?” Bad question: “Do you find our product useful?” (leading, yes/no)
Adjustments: Include warm-up questions to build rapport. Add probing follow-ups for each main question. Avoid any questions that assume the product is at fault — the issue could be internal adoption, training gaps, or misaligned expectations.
Type of Output: Interview guide with timing estimates per section, 4 main question blocks with 2–3 probes each, and a closing section.
Extras: Include an ethical note about informed consent and data handling. Remind interviewers that AI-generated guides should be piloted with a test interview before full deployment. All findings should be verified against actual usage data.
Enterprise Churn Discovery Interview (30 min)
Pre-Interview: Confirm informed consent. Explain recording policy and how data will be used. Remind participant there are no right or wrong answers.
Warm-Up (3 min)
“Tell me about your role and what a typical week looks like for your team.”
Block 1: Adoption Context (7 min)
“Take me back to when your team first started using our product. What problem were you hoping it would solve?”
Probe: “How did the reality compare to that expectation in the first month?”...
Note: Pilot this guide with one internal test interview before using with actual customers. Cross-reference interview findings against product analytics data.
When to Use CREATE
Best for tasks that benefit from persona adoption, examples, and iterative refinement
Perfect For
When the output must match a particular brand voice, professional persona, or writing style — the Character component anchors the AI in the right identity from the start.
When you need the AI to match a specific template or output structure — providing an Example of the desired format is far more effective than describing it in words alone.
Tasks where the first draft is a starting point, not a final product. The Adjustments step formalizes the refinement loop that experienced prompt engineers use instinctively.
Marketing copy, training materials, documentation, and any task where multiple constraints (voice, format, audience, guardrails) must be satisfied simultaneously.
Skip It When
Simple lookups do not need a persona, examples, or iterative refinement. Asking “What is the boiling point of water?” with six CREATE components adds overhead without value.
Mathematical proofs, logic problems, or code debugging where the challenge is analytical accuracy. Chain-of-Thought or Tree-of-Thought techniques are better suited for reasoning tasks.
Early ideation where you want the AI to surprise you. Specifying all six components too early can constrain the creative space before you know what direction to take.
Use Cases
Where CREATE delivers the most value
Marketing Copywriting
Generate brand-voice-consistent copy across product pages, email campaigns, and social media — with examples ensuring tone consistency and extras providing legal guardrails.
Educational Content
Create lesson plans, explanations, and study guides where the Character sets the teaching approach and Examples demonstrate the desired explanation style.
Technical Documentation
Produce API documentation, user guides, and onboarding materials where format consistency and audience-appropriate language are critical to adoption.
HR and People Operations
Build performance review templates, job descriptions, and internal communications where voice consistency and legal sensitivity are essential.
Customer Communication
Draft support responses, onboarding sequences, and feedback requests that match your brand voice while addressing specific customer segments and concerns.
Training and Workshops
Design training scenarios, role-play scripts, and workshop materials where the Character component creates realistic professional simulations for learners.
Where CREATE Fits
CREATE bridges basic structured prompting and full creative briefing
Unlike frameworks that treat prompting as a single step, CREATE explicitly builds in iteration through its Adjustments component. After seeing the AI’s first response, you can add new Adjustments without rewriting the entire prompt — making CREATE especially powerful in conversational AI interfaces where multi-turn refinement is the norm. Treat each round of Adjustments as a focused edit pass, not a fresh start. And always verify the final output before using it in production.
Related Techniques & Frameworks
Explore complementary approaches to structured prompting
Build Your CREATE Prompt
Structure your next prompt with all six CREATE components or find the right framework for your specific task.