How to Turn a Vague Brief Into an AI-Ready Specification
A question-driven method for clarifying goals, evidence, constraints, output contracts, and acceptance tests before using AI.
Find the decision hidden inside the request
Requests such as “make the website better” or “write about our product” hide a decision. Better for whom, measured by what behavior, under which constraints, and using which evidence?
Rewrite the request as an outcome with an owner and a next action. This immediately separates useful context from decorative detail.
Rewrite the pricing-page hero for US-based freelance designers who compare us with desktop tools. The goal is to increase clicks to the free browser demo. Use only the supplied privacy and format-support facts.
Ask only questions that change the output
A long intake form creates friction without guaranteeing clarity. Ask questions whose answers would materially change the recommendation, structure, or risk.
The highest-value questions usually concern audience, desired action, authoritative sources, fixed constraints, prohibited claims, and the delivery format.
- Who will use or approve the result?
- What should happen after they read or run it?
- Which facts are authoritative?
- What cannot change or must not be claimed?
- How will the result be reviewed or tested?
Separate facts, assumptions, and choices
Models often turn an unstated assumption into confident prose. Label confirmed facts, working assumptions, and open choices separately.
Tell the assistant what to do when a required fact is absent: ask a question, use a labeled placeholder, or present alternatives. Silence invites invention.
Finish with an output contract and acceptance test
Specify the shape of the answer only as far as downstream work requires. A human decision may need a table and recommendation; software integration may need a schema; a coding task needs runnable tests.
Before submission, read the specification as if you had no project memory. If an external collaborator could not distinguish success from failure, the AI cannot reliably do so either.
- Required sections or fields.
- Length, tone, channel, or compatibility limits.
- Evidence and citation behavior.
- A checklist that makes success observable.
Turn the method into a usable prompt
Enter a rough idea and PromptSmith will add structure, constraints, and an output format.
Optimize a prompt free →