A task lacks a time range or subject: how do you make AI ask when needed instead of guessing?

Decide whether missing information changes the result

Examples include “organize this month's data” without a time range or “edit this file” without a target version. Identify the possible outcomes: a calendar month differs from the past thirty days. A default style may be acceptable for wording, but clarify details that determine the target or statistical scope.

Check existing context before asking again

Separate required task fields from optional preferences and check whether the context already provides each answer. Do not re-ask information just supplied or treat an inference as confirmed. In multi-turn work, distinguish a supplement to the current task from an explicit replacement so original constraints survive clarification.

Ask only for details that affect execution

State the missing information and its effect on the result, preferably with a few mutually exclusive choices. Once answered, save the confirmed conditions and continue authorized work. Independent read-only organization need not wait for an unrelated preference, but elapsed waiting time is not consent for a critical target or permission choice.

Verify that the answer is used and work continues

Test incomplete and complete requests. The former should trigger focused questions; the latter should proceed without repeating known conditions. Verify that the user's answer changes subsequent processing rather than eliciting only an acknowledgment. The final time range, target or version should match the deliverable.

This page designs clarification behavior and authorizes no business action. Applications may not retain state across turns, requiring confirmed conditions in the current request. Assumptions should remain correctable; uncertain accounts, recipients or resources must not become stated facts. If essential information is unavailable, report incomplete work instead of pretending success.