Improve the prompt without changing the goal
Clarify the instruction, preserve its meaning and make every proposed change explainable.
Keep the requirements that already matter
Refinement should resolve ambiguity without replacing the user’s objective. Preserve required facts, exclusions, audience and output constraints. If the original contains an unresolved choice, ask about it rather than choosing silently.
Compare the original and revised prompt side by side. The revision should address a specific weakness and still ask for the same work.
Refine a research request with a missing time boundary
The revision should expose what it cannot safely infer.
Refine this request without choosing an unstated market or date range. Identify the geography, period, audience and source requirements that need clarification. Propose a concise prompt that distinguishes observed data from estimates and requires publication dates. Preserve the request for a short report; do not expand it into an unrelated strategy project.
Compare the results after the wording changes
A cleaner prompt may still perform poorly on the target model or sources. Keep the original as a baseline, test representative inputs and retain the revision only when the results support it.
Common questions
Can refinement add unsupported details?
It should not. Treat new facts, scope or constraints as proposed choices requiring review, rather than silently inserting them into the instruction.
Sources and review
Reviewed