How to Rewrite a Weak ChatGPT Prompt in 60 Seconds

You already have the prompt. It sort of works. The output is fine but generic, and you're not sure what to change. Here's the 60-second fix — the same four-move rewrite whether you're prompting ChatGPT, Claude, or Gemini.

Start from what you already wrote

Don't throw the prompt away. Your original contains the one thing the rewrite needs: your actual intent. Everything else is scaffolding you're about to add around it.

Take this one:

write me a cold email for my agency

The intent is clear. What's missing is everything that makes the answer yours.

Move 1: Assign the role (10 seconds)

Open with who the AI should be. Be specific and senior — the role sets the ceiling on the answer.

Act as a cold-outreach specialist whose first emails get above a 10% reply rate.

Not "act as a marketer." A marketer writes marketing. A specialist with a stated benchmark writes to hit the benchmark.

Move 2: Paste in your situation (20 seconds)

This is the move people skip, and it's the one that matters most. The model can only tailor to details you give it.

I run a [service] agency targeting [ideal client]. Their biggest pain is [pain]. My differentiator is [what makes you different].

Use bracketed placeholders for anything that changes between runs — that's what turns a one-off prompt into a reusable one.

Move 3: Set the constraints (15 seconds)

Tell it what to optimize for and what to avoid. Unconstrained, models drift toward the safe average of their training data — which is exactly the generic tone you're trying to escape.

Keep it under 90 words. One clear ask. No hype, no "I hope this finds you well," no flattery. Avoid anything that reads like a template.

The "avoid" list is doing more work than the "do" list here.

Move 4: Specify the output (15 seconds)

Say exactly what you want back, including structure and quantity.

Give me 3 variants: one referencing something specific about their business, one leading with a free insight, one asking a genuine question. For each, add one line explaining the psychology behind it.

The rewrite, assembled

Act as a cold-outreach specialist whose first emails get above a 10% reply rate. I run a [service] agency targeting [ideal client]; their biggest pain is [pain]; my differentiator is [differentiator]. Write 3 opener variants: one referencing something specific about their business, one leading with a free insight, one asking a genuine question. Each under 90 words, one clear ask, no hype or flattery, nothing that reads like a template. For each, add one line on the psychology behind it.

Same request. The first version gets you a cold email that could belong to anyone. This one gets you three you can send today.

Why this works on every model

These four moves aren't a ChatGPT trick. Every large language model is a next-token predictor conditioned on your input — vague input means it falls back to the statistical average of everything it has seen, which reads as generic because it literally is. Role, context, constraints, and format each narrow that distribution toward what you specifically want.

Models will keep getting better at guessing your intent. They still can't guess your budget, your audience, or the three approaches you already tried and hated.

Two habits worth building

Keep your rewrites. The prompt you just engineered is reusable — only the bracketed parts change next time. A folder of ten engineered prompts beats re-improvising every session.

Check before you commit. If you're not sure whether the rewrite is actually better, paste both into the Prompt Grader and compare scores. Watching a 13 become an 89 teaches the method faster than any article, including this one.

And if you'd rather start from a strong prompt than fix a weak one, the Prompt Builder assembles one from a few questions — tuned to the model you're using and scored before you copy it.