A better AI prompt starts with a clear definition of done
Turn a vague request into a short brief that makes the output easier to review.
If you ask an assistant to improve a paragraph, it has to guess what improvement means. Shorter? More formal? Easier to read? More persuasive? A useful prompt removes those guesses before adding elaborate instructions.
Try a five-part brief: task, audience, approved facts, limits and output format. For example: rewrite these opening hours for first-time visitors; keep every time unchanged; use plain English; do not invent holiday arrangements; return a headline and three short sentences. This is an editorial workflow example, not a guarantee that a model will follow every instruction.
Add an acceptance test you can perform yourself. In this example, compare the hours character by character with the original. Read the text as a visitor who does not know the business. Check that no sentence suggests a service you did not supply in the brief.
The most helpful follow-up is usually specific. Instead of asking for a better result, identify the defect: the draft says Sunday delivery is available, but the supplied facts do not say that. Request a revision based only on those facts. Keep the original information beside the final text so the review remains grounded.
NIST's generative AI guidance discusses the risk of plausible but incorrect output. A detailed prompt can support a clearer workflow, but it does not replace verification. For repeated tasks, save a brief template and a review checklist together. The template makes the request consistent; the checklist establishes whether the answer is ready to use.