Interview prompting means having the AI interview you until the task is clear, instead of pretending you gave it enough context. The line we use in a prompt: “ask me any questions one at a time, to gain as much context as you need to succeed at this task.” The “one at a time” part matters, because twelve questions at once means you skim, answer badly, or quit.
A contractor remodeling a room shows the difference. A bad one hears “make it nicer” and starts swinging a hammer. A good one asks first: What’s the budget? Who uses the room? What does done look like?
Why you care
Most bad AI output comes from the model missing real constraints you forgot to mention: who the audience is, or that the client is sensitive about pricing. Interview prompting works best on fuzzy work (a proposal, a workflow, a client email) where the missing context lives in your head; the AI asks, you answer, and the context gets built through conversation. It’s the I in the CRIT prompt framework, and it keeps you in the human in the loop role where your judgment matters most: defining the work.