Most disappointing AI output comes from under-specified prompts, and the fix is mechanical. These 10 patterns work in ChatGPT, Claude, Gemini and Copilot. Copy the template, fill the brackets, stack patterns freely.
The 10 patterns
- Role.
You are a [senior email copywriter for e-commerce brands].— sets expertise and tone in one line. - Context dump.
Context: [audience, product, goal, constraints].— the model cannot use what it does not know; give it everything relevant, ugly formatting is fine. - Output format.
Answer as: [a 5-row markdown table with columns X, Y].— kills rambling instantly. - Show an example.
Here is one example of the style I want: […]. Match it.— one example beats three paragraphs of description. - Constraints.
Max 120 words. No exclamation marks. Grade-7 reading level.— boundaries produce discipline. - Think step-by-step.
Reason through this step by step before giving your final answer.— for anything with logic or numbers. - Critique-then-rewrite.
Now critique your answer against [criteria] and rewrite it stronger.— the single highest-leverage second message. - Audience lens.
Rewrite this for [a busy founder who skims].— same content, different reader, different words. - Iterate narrow. Change one thing per follow-up (“same, but shorter hooks”) instead of re-prompting from scratch.
- Demand honesty.
If you are not sure, say so. Mark any claim that needs a source with [verify].— cheap insurance against confident nonsense.
FAQ
Which AI tool do these work with?
All major chat models — ChatGPT, Claude, Gemini, Copilot. The patterns are model-agnostic because they fix the prompt, not the model.
What is the fastest way to level up from here?
Stack three patterns per prompt: Role + Context + Format solves most everyday tasks. Add Critique-then-rewrite when quality matters.