You’ve had this experience. You ask AI a reasonable question. You get a reasonable answer. It’s coherent. Well-structured. Even insightful.

And somehow, it’s not the answer to your question. It’s the answer to a question shaped like yours.

The reason isn’t what you think.

There’s a story about AI that’s so common you’ve probably absorbed it without noticing. The story says: you give AI a question, AI gives you an answer. One thing in, one thing out. The better your question, the better the answer.

The story is incomplete, by exactly the part that matters.

AI needs two things from you. Not one. Two.

The first is obvious: the gap — what’s missing, what you want to know, what’s unresolved. That’s the question you typed.

The second is the part nobody talks about: the field — the larger structure your question belongs to. The whole that your gap is a hole in.

AI fills the gap. You define the space the gap exists in.

If you don’t define the field, AI picks one. It has to — completion can’t happen without a whole to complete. So AI reaches for the most statistically likely field given your wording. Which is often a default, generic, average field. Which is how you get answers that are confidently, competently wrong.

This is why prompt engineering feels exhausting. Most of the advice is about refining the gap — add more detail, be specific, give examples. And sometimes that works. But a lot of the time, the gap is fine. What’s missing is the field.

The word “context” doesn’t capture it. Context sounds like surrounding information — who you are, what the project is, extra background. A field is different. A field is the kind of whole you’re treating your situation as. It tells the pieces how to relate. Same pieces, different field, different universe.

Take a real question: “Why can’t I focus?”

Ask AI without specifying the field. You’ll get a pretty standard answer. Probably: reduce distractions, put your phone away, try Pomodoro, check your sleep, maybe see a doctor about ADHD.

Generic. Not wrong. Not useful either.

Now specify the field.

Environment. “Why can’t I focus in my current workspace?” Now the answer is about physical setup, interruptions, lighting, noise, the chair you sit in. Practical. Immediately actionable.

Motivation. “Why can’t I focus on work I used to care about?” Now the answer shifts entirely. It asks whether the work is still yours, whether the stakes are real to you anymore, whether you’re in the right role. Different world.

Cognition. “Why can’t I focus the way I used to — is something wrong with me?” Now the answer goes to sleep architecture, attention patterns, possible medical things worth investigating. A third different world.

Systems. “Why can’t I focus when I have twenty unfinished projects open?” Now the answer is about prioritization, closing loops, making a “done” bar visible. Operational, not personal.

Same question. Same gap. Four radically different answers — each coherent, each useful, each useless if you’re in the wrong field.

The work you do in the thirty seconds before you type isn’t preamble. It’s the thing that decides what kind of answer AI is even capable of giving you.

Before your next real AI question, do one thing. Decide what kind of problem you’re holding.

A short list to pick from:

Pick one. State it out loud or in your head. Then ask your question inside it.

If none of the options fit, that’s useful information — you don’t know what kind of problem you have yet. The move isn’t to ask AI anyway. It’s to sit with the not-knowing for a moment. Sometimes the field clarifies itself if you wait. Sometimes it doesn’t, and you have to try a few and see which answer reorganizes the situation.

If AI gives you a coherent-but-useless answer, don’t rewrite the question. Change the field. Watch how differently it responds.

Here’s what stays with me.

There’s usually more than one field that fits a situation. Sometimes they complement each other — communication AND trust, efficiency AND motivation. Sometimes they conflict. “This is a strategy problem” and “this is a people problem” don’t just give different answers; they point at different universes.

Which makes field selection feel like a choice. And a choice has stakes.

When you specify a field, you get coherent answers inside that field. That’s the gift. But you also foreclose the other fields. You won’t hear what your situation looks like through a different structure unless you deliberately go back and ask.

So: does the field you pick shape reality, or recognize what’s there? If a situation can equally well be framed as a trust problem or a systems problem, is one of those actually right? Or does the framing itself decide which one becomes real?

I don’t know. But I’ve started running the important questions through two fields before I commit. The disagreements are where the real information is.