Find the gaps in the instruction
Check whether the prompt gives the model enough direction to do the intended job.
Review the request before blaming the answer
A prompt may omit the audience, conflict with its own constraints or request evidence the tool cannot access. Prompt Score helps review the instruction so you can identify what to clarify before using or reusing it.
- Does the task have a recognizable finish?
- Are the relevant sources actually available?
- Are conflicting constraints resolved?
- Does the output format match how the result will be used?
Review what the prompt revision fixes
Explain the effect of an edit instead of rewarding extra length.
Review this instruction for ambiguity. Identify the missing dataset, business question, audience and output format. Propose the smallest revision that makes the analysis checkable. Do not invent a dataset or desired conclusion. Explain which ambiguity each edit resolves, and leave unresolved business decisions as questions.
Prompt quality and answer accuracy are different
A high score is not a guarantee of factual correctness, task success or better revenue. Run the prompt against representative inputs and review the resulting answers. Avoid treating a rubric as an independently calibrated probability of success.
Common questions
Is a longer prompt always better?
No. Extra instructions can introduce contradictions or irrelevant detail. Improve clarity and relevant context, then test whether the change helps.
Sources and review
Reviewed