Prompting Is Just Asking Well

There is a genre of advice that treats prompting as a secret dialect — magic words, role incantations, a prefix that unlocks the good model. I have tested a great deal of it. Most of it is superstition built on a single lucky afternoon.

What survives testing is much less exciting. The people who get consistently better output are the people who ask better questions, and they were mostly good at that before any of this arrived. So I stopped teaching prompting and started teaching inquiry, borrowed from three professions that have to ask well or fail publicly.

The four moves

1. Say the situation before the request. A journalist does not open with the question; they open with why they are calling. Two sentences of context change the answer more than any amount of rephrasing the request itself.
2. Ask for the thing under the thing. Therapists do this relentlessly. "What would have to be true for this to be a good idea?" gets you the assumptions; "is this a good idea" gets you a paragraph of agreement.
3. Ask again, differently. This is the eight-year-old's move and it is badly underrated. The second and third versions of a question are where the interesting material is, because you have learned something from the first answer and can aim better.
4. Ask what you have not asked. "What is the question I should be asking here?" It fails maybe half the time. The other half it hands you the thing you were circling.

None of those require you to know anything about how a model works. They require you to be willing to be specific in public, which is a much rarer skill than it sounds.

A vague question protects you from being wrong about anything in particular. That is exactly why it is comfortable, and exactly why it is useless.

A worked example

A department head asked a model to "help write our AI policy." What came back was a policy — generic, reasonable, and useless to her, because it had no idea that half her faculty were furious and the other half had been quietly using these tools since March.

We rewrote the opening as situation first: who is in the room, what the last meeting sounded like, what the superintendent will and will not sign, what she personally believes and is not sure she can defend. Then the same request. The second answer was not magic, but it was about her school, and it disagreed with her twice, which is the part she found useful.

That is the whole method. Say more about where you stand, ask for the layer underneath, ask again, then ask what you missed. It is interviewing. We have known how to do it for a century.