A repeatable method for turning vague requests into clear, testable AI work instructions.
Start with success
Describe what a useful answer lets someone decide or do. A measurable target is stronger than adjectives such as good, professional, or detailed.
Supply decision-relevant context
Include audience, market, source boundaries, constraints, and important definitions. Remove background that cannot change the answer.
Define the output contract
Name the sections, order, format, length, evidence expectations, and what to do when information is missing. A small example often resolves ambiguity better than another paragraph of rules.
Iterate with one change
Keep a fixed test input and alter one prompt component at a time. Compare completeness, factual support, usefulness, and failure modes before publishing a new version.