Write a brief that gives AI something concrete to finish

You do not need a long prompt to get useful work from an AI platform. You need a clear task. A good brief explains what should exist when the work is done, what information may be used, and how you will judge the result. That is the same foundation you would give a capable colleague.

Describe the finished object

Ask for a document, comparison, table, draft, or list of unresolved questions. Name the audience and the decision it should support. “Create a one-page handoff for the plumber” is easier to evaluate than “be an expert property manager.” A role description can add context, but it does not replace the actual assignment.

Provide source material and identify its status. A signed agreement, a draft note, and an unverified message carry different weight. Ask the AI to preserve that distinction and flag gaps. Do not instruct it to fill every blank if doing so would require inventing a fact.

A four-part brief

• Outcome: the useful deliverable and intended reader.
• Sources: the permitted material and relevant version or period.
• Boundaries: what may be suggested, drafted, read, or changed.
• Check: the criteria for accepting the result.

For example: “Using these synthetic notes, create a repair handoff for review. Include the reported issue, location, prior work, and unanswered questions. Do not diagnose the cause or contact a provider. I will check each fact against the source.” This is concise but gives the task a real shape.

Improve the brief through feedback

When the result misses something, give the specific correction and explain why it matters. If the output is too long, name the parts to keep rather than asking for arbitrary brevity. If the task changes, update the outcome and constraints so the AI does not optimize for an abandoned version.

Save the useful pattern

Keep briefs that consistently produce useful, checkable work. Replace source dates and context each time rather than recycling stale facts. A reusable brief is a working aid, not a guarantee of correctness or permission. The final test is still whether the actual deliverable meets the task and whether any external action was properly authorized and verified.