Ask the AI what it needs

A context-gathering method in which you ask an assistant to identify the examples, rules, records, or decisions it still needs for a defined task. Its answer is a candidate checklist that you verify before supplying anything.

Ask the AI what it needs when you know the outcome but are unsure which context will help. Define the task first, then ask the assistant to list the missing examples, rules, records, and decisions it would use. For customer-support replies, it might request past replies, the current return policy, and a product list. Check that list against the real workflow, remove irrelevant or sensitive material, and supply the smallest useful set. The assistant's request can expose a gap you missed, but it can also omit an important source or ask for data it does not need.

Builder example

This question gives context gathering a concrete starting point. A meeting-summary assistant might ask for the expected format, participant names, and the project's open decisions. You can then test a summary with those inputs and compare it with a summary that lacks them. The comparison shows which context improves the result; the model's initial request alone does not establish that every item is necessary.

Common confusion: The assistant is proposing inputs, not setting the data policy. Its list does not grant access to private files, prove that an item is relevant, or replace your application's permission rules. Connected agents may be able to fetch allowed sources, while an unconnected chat can use only what the interface supplies.