Characterise, Request, Examples, Adjust, Test, Evaluate — A six-part, iteration-friendly framework built for getting the details right.
CREATE is designed for tasks you expect to refine rather than get right on the first try — long-form content, technical writing, anything where the first draft is a starting point. It builds revision into the framework itself instead of treating it as an afterthought.
Define the role, tone, and expertise the AI should bring to the task.
State the core task clearly and specifically.
Show the style, structure, or reference material to draw from.
Note any constraints or tweaks to apply on top of the base request.
Describe how the output should be checked or validated before it's considered finished.
State the criteria that make the final output good enough to ship.
Bad prompt:
“Write a technical guide about API rate limiting.”
Structured with CREATE:
Characterise: You are a backend engineer writing documentation for other engineers. Request: Write a guide explaining how to implement API rate limiting. Examples: Follow the structure of Stripe's API docs — short paragraphs, code snippets, no marketing language. Adjust: Focus on token bucket and sliding window approaches only; skip anything cloud-provider-specific. Test: A mid-level backend engineer should be able to implement a basic version after reading it once. Evaluate: Success means the guide has runnable code, not just conceptual explanation.
A structured technical guide covering token bucket and sliding window rate limiting, with short explanatory paragraphs and runnable code examples in the style of vendor API documentation, avoiding marketing language.
CREATE's structure is particularly effective with Claude for longer technical or educational content, and works well with ChatGPT when you plan to iterate across multiple turns.