How to Get Exactly What You Want Even When You Don't Know What to Ask

You don't know what you don't know. That’s fine most of the time. But it’s a problem when you’re using AI on something you can’t already easily describe in a paragraph: scoping a new product, planning around constraints, drafting a pitch you don’t have the angle on yet.

A prompt can only ask for what you’ve already thought of. Everything you forgot to mention, every question you didn’t think to ask, gets left out. You get back an answer shaped around what you said, not what you actually wanted.

The fix is simple: don’t write the prompt alone. Tell your AI what you’re trying to do and let it interview you. The questions bring out ideas you didn’t think to include and cover ground you didn’t know to cover.

Give your AI a goal in one sentence: what you're trying to build, decide, plan, or write. Tell it to interview you (I prefer one question at a time), and to ask twice as many questions as it thinks it needs.

The twice-as-many part matters because AI tends to jump to conclusions. By question three, it's already guessing what you want and starting to steer toward an answer. The extra questions force it to keep asking instead of jumping ahead.

The interview usually settles into six to twelve questions. When there's a clear set of options, your AI will offer multiple choice. You can click through three options in twenty seconds.

Try it yourself. Open whichever AI you use. Pick a task you can't already write a paragraph about. Paste in this prompt and replace the bracketed line with your goal:

When the AI has enough, it'll tell you. Add anything else that comes to mind, then let it put together the first version.

Two things happen when you let your AI run the questions instead of giving instructions.

It walks into the conversation with the entire internet in its head. If you're building a website, it's already read about a few thousand of them. It knows the questions a good web designer would ask before starting work. You might not. Most of those questions would be sitting in your blind spot. When the AI asks them, you suddenly have something to answer.

Being asked a question also forces a different kind of thinking than writing a prompt does. When you type a prompt, you're filtering. You compress your thoughts into what you think is relevant. When you answer a question, you talk. You ramble. You hear yourself say something you didn't plan to say. That accidental sentence is often something that matters.

A “good enough” prompt won’t get you a finished draft. You get out of AI what you put in, and an interview is the version of putting in that doesn’t ask you to already know what to put in.

If the outcome is important to you, don’t accept what the AI returns as a finished product. It’s either a draft of a prompt you’ll run, or a starting point for the conversation that gets you to the deliverable.

You already know what good looks like, since you pinned that down during the interview. Hold the first version up to it. Where does it match? Where does it miss? That’s your edit list. Without that bar, it’s easy to read something that sounds right and call it done.

Read what came back the way you’d read something from a junior teammate. Push back on what doesn’t fit. Point at what works and ask for more.

What comes next is sharpening that first version into something you’d actually send, ship, or act on.

I used to spend a lot of time writing prompts, and I thought I was doing it right. At the time, I didn't realize how much I was leaving out. Now I tell my AI what I want and let it pull the rest out of me. The output is sharper, I’m doing less of the work, and the questions teach me things about my own task I didn't know yet.

me+machine

Your turn: What task are you going to interview yourself on? One sentence. Hit reply.