Turn a Long AI Prompt Into a Clear Task Contract

A long prompt can still leave the most important instruction unclear. A task contract is a short statement of the input, the permitted transformation, the required output and the checks that define completion. Use it when repeated revisions keep drifting from your goal.
Name the transformation
Compare “Improve this” with “Rewrite this public event announcement in plain English while preserving the date, venue and registration link.” The second instruction tells the assistant what may change and what must survive. Add the intended reader and a practical length limit if they matter.
For a fictional notice, list the facts separately: Saturday, 10:00, West Hall, registration required. Supply the actual notice below a clear input marker. Do not include real personal information merely to make the example feel realistic.
Specify the output shape
Ask for a headline, a two-paragraph announcement and a short checklist of preserved facts. If a fact is missing from the input, require a question or an unknown marker rather than a completion. Explain whether suggestions should appear inside the copy or in a separate editor note.
A useful prompt ending is: “Before returning the final copy, check every listed fact against the input. Do not replace the registration URL or invent accessibility arrangements.” This describes a check, but you must still perform your own comparison.
Test with one controlled revision
Change one requirement at a time. If the tone is wrong, adjust the tone instruction without also changing the audience, length and structure. Save the previous version so you can tell what the new instruction affected.
Accept the output only after the facts, links and requested format match. A task contract reduces ambiguity; it does not make an assistant's self-check reliable enough to replace editorial review.
Comments