When a request is ambiguous, a model usually guesses. Sometimes you'd rather it asked.
When Clarification Helps
- Requests with several reasonable interpretations.
- Missing critical information: audience, budget, deadline, constraints.
- High-cost mistakes if the guess is wrong.
How to Prompt for It
"Before starting, ask me up to three questions about anything you need to know to do this well. Wait for my answers."
For applications: "If the request is ambiguous or missing required details, ask one concise clarifying question instead of answering."
Define What's Required
Tell the model which details are essential ("destination and travel dates are required") so it asks about those rather than trivia.
Limit the Questions
Too many questions frustrate users. Ask for the most important one or two, or propose sensible defaults: "I'll assume a 30-minute presentation for executives unless you say otherwise."
Offer Assumptions as an Alternative
"State any assumptions you make at the top of your answer" often works better than questions for low-stakes tasks, because the user can correct them.
Test Both Paths
Evaluate with ambiguous and clear requests: the model should ask for clarification only when it's genuinely needed.