The problem is not brevity
A one-line request can contain a strong idea. The failure begins when a generator simply expands the wording without settling the decisions hidden inside it. The destination AI then has to rediscover the audience, operating environment, boundaries and definition of done.
A complete prompt does not need to predict every implementation detail. It does need to close the questions that would materially change the result.
Resolve consequential decisions
Ask only questions whose answers change architecture, cost, risk or user experience. Present a recommended answer first, explain the trade-off, and retain a custom-answer path. This keeps the interview fast without pretending every project is identical.
Useful decision groups include the intended user, delivery surface, data sensitivity, preferred stack, research policy, deployment environment and acceptance checks.
End with execution architecture
The final prompt should tell the destination AI what to build, in what phases, which boundaries cannot be crossed, how each phase is verified and when stable work is frozen. It should not restart discovery with another unanswered questionnaire.