An unsatisfying answer does not tell you which part of the task went wrong. Before asking for a better version, identify the mismatch you can actually check. Then change the instruction or input that addresses it.
1. Name the problem
Missing information and a writing-style problem need different fixes. A more forceful instruction cannot supply a source the AI has not received or checked. If the task depends on current facts, provide suitable material or use an available research tool.
- Too general: add the reader, decision and relevant task facts.
- Wrong format: supply the intended structure and a short example.
- Unsupported facts: provide sources and require uncertain claims to be marked.
- Wrong level of detail: define what to keep, what to cut and a length target.
2. Make one concrete correction
Point to the passage that fails your requirement. State the replacement criterion and identify what to keep. Give the next draft a clear target.
This is not good. Make it more professional.
The opening repeats the background and hides the request. Keep the project facts, move the decision we need into the first sentence, and put the two available options in bullets. Remove the generic introduction. Do not add facts. Keep it under 150 words.
Illustrative revision requests: identify what to keep and give the next draft a checkable target.
3. Copy this repair prompt
Use the actual draft and a few checks. Ask for the revision and a brief change summary so you can compare them easily.
Original task: [paste the request].
Current draft: [paste the result].
What is wrong: [one specific mismatch, with an example].
Keep: [useful content or structure].
Change: [a checkable requirement].
Sources and facts: [the material the revision may use].
Revise the draft. Do not invent evidence or change supplied facts.
After the revision, briefly list the changes and any requirement that remains unmet.4. Test the revision before saving a rule
Compare both drafts against the same checklist: correct facts, required content and usable format. Keep the task and source material unchanged while testing one instruction. Try it on another task before saving it as a standing rule.
Check citations against original sources. Test code and calculations. AI self-reviews can miss mistakes; a claim that everything is correct is not enough.
When should I try a different prompt?
Choose a different method when the current one solves the wrong problem. A feedback mentor can help review an assignment; a paper-analysis prompt can organize a critical reading. Neither replaces missing evidence or your judgment. Model behavior and outputs can vary, so keep testing the result you need.