The interview technique: how to get far better answers from AI

One of the most effective ways to use AI has nothing to do with asking it questions. It’s letting it ask you.

I use AI as a critical thought partner more than anything else, and the single technique that has made the biggest difference to the quality of what I get back is the interview. Instead of firing a question at it and taking whatever comes back, I give it a role and tell it to interview me first, one question at a time, before it answers. The difference in the responses is significant. This edition is about how that works, why it beats a straight question every time, and the second habit that matters just as much: never accepting the first reply.

Why a straight question gives you a mediocre answer

When you ask AI a plain question, it has almost nothing to work with except the words in front of it. It doesn’t know your situation, your constraints, or what you’ve already tried. So it does the only thing it can. It gives you the average answer, the one that would suit almost anyone and therefore suits no one particularly well.

Picture a managing director typing “how do I improve staff retention” into ChatGPT. Back comes a tidy list: competitive pay, clear progression, recognition, flexible working. All true. All useless, because it’s the same list their competitor two miles away just received, and it says nothing about the fact that their problem is one team, one manager, and a pay structure they inherited and can’t easily change.

There are a few reasons this happens.

• AI (generative AI like Claude, ChatGPT, Gemini, CoPilot) is a prediction engine. It works by predicting the next most likely word based on the enormous amount of text it was trained on. Ask it something generic and it returns the statistically generic response.
• It has no context you haven’t given it. It cannot see your business, your team, your numbers, or the specific decision keeping you up at night. Without that, it fills the gap with generalities.
• Most people prompt it the way they were taught to search. We spent two decades learning to type short queries into Google. That habit produces short, shallow exchanges with a tool built for something much richer.

Used like a search box, AI stays a search box. The technique below is how you turn it into a thought partner.

The shift: from asking to being asked

The most useful instruction you can give AI is to have it interview you before it answers.

Instead of demanding a verdict, you give the AI a role and put it in the questioner’s seat. It asks you one question at a time, each answer shaping the next, until it understands your situation well enough to say something genuinely useful.

You stop being a person collecting an answer and become a person being helped to think.

Take that same retention problem. Here’s the instruction rewritten as an interview:

“You are an experienced HR director who has turned around retention in small professional firms. I’m losing good people and I want to understand why before I throw money at it. Interview me one question at a time, asking at least five questions, to understand where the losses are concentrated and what’s driving them. Then give me your assessment and the two things I should act on first.”

Now the exchange looks completely different. Its first question might be: “Are your leavers spread evenly across the business, or concentrated in one team or level?” You answer that it’s mostly one team. Next question: “What’s changed in that team in the last twelve months?” You realise, as you type, that the answer is a new manager. The AI hadn’t diagnosed anything yet, but you already had.

The mechanics behind this are simple. A strong version of the instruction has four parts:

• Context. Describe the situation and give plenty of detail.
• Role. Tell the AI who to be. “Act as an experienced operations director.” “Act as a seasoned negotiator.”
• Interview. Ask it to interview you one question at a time, and set a floor of at least three to five questions.
• Task. Tell it what to produce once it understands your situation.

Two practical notes matter more than they look. Tell it to ask one question at a time, or it will fire ten at once and you’ll abandon the whole thing. And set the number of questions deliberately: fewer than three and it barely learns anything before answering, so five is a sensible floor for any decision of real weight. One at a time, at least five deep, keeps it a proper conversation rather than a form.

What you can actually use this for

The retention example is one case, but the technique works anywhere you’d benefit from thinking something through with a sharp colleague, mentor or even a competitor. A few of the situations where it is most effective.

• Making a difficult decision, where you talk through the options and have the AI pressure-test each one before you commit
• Preparing for a hard conversation, by having it interview you about the person and the history, then draft something that will actually land
• Writing a job description, where it questions you on what the role really needs before producing a draft, so you’re not just listing tasks
• Preparing for a meeting where you expect pushback, by having it interview you about who’s in the room and what they care about, then rehearse the likely objections
• Drafting a pitch or an important message, where it pulls the key points out of you rather than you staring at a blank page
• Building or reviewing a plan, where it interviews you on your goal and assumptions, then tells you where the plan is strong and where it’s thin

That last one is worth dwelling on. When you’re stuck for a use, the interview itself finds one. Tell the AI you’re not sure how it can best help this week and ask it to interview you until it spots something. It’s a good way to break the blank-page feeling that stops most people using AI properly at all.

Why the AI interview produces a better answer

Three things happen when you let the AI question you, and each one improves the output.

It pulls context out of your head that you were carrying loosely. You know far more about your situation than you consciously articulate. Watch what happened above: you didn’t get the answer from the AI, you got it from being asked the right question in the right order. The interview forces you to lay out what you already half-knew.

It grounds the answer in your reality rather than the average. Once the AI understands your actual constraints, its response is about your situation, not situations in general. Consider someone deciding whether to raise their fees. Ask flatly and you get textbook pricing theory. Let it interview you across five or six questions and it learns that the real fear isn’t the number, it’s one long-standing client who accounts for forty percent of the income. Now the advice is about managing that dependency, which is the actual problem, rather than about pricing, which was only the symptom.

It challenges you. A good interviewer asks the question you were avoiding. Imagine someone convinced they need to hire a salesperson. Told to act as a sceptical board member and interview them, the AI asks: “Before we discuss hiring, how do you know sales capacity is the constraint, rather than your conversion rate on the leads you already get?” That single question can save a year and a salary. It’s the kind of thing you don’t see because you’re inside the jar and can’t read the label.

This is the same reason a strong mentor beats a good textbook. The textbook has answers. The mentor asks what you’ve actually tried, and in answering, you find the way forward yourself.

Never accept the first reply

Here’s the habit that separates people who get real value from AI from people who get a slightly better search engine. When the AI finally gives you its answer, don’t take it. Push back on it.

The first response is a draft, not a verdict. It’s usually somewhere between half and two-thirds of what you actually need. Your job is to make it earn its place, and the fastest way to do that is to turn the AI’s critical eye onto its own work.

A few ways to push:

• Ask it to critique itself. “Now act as a sceptic reviewing that answer. Where is it weakest? What did you assume that might not hold? What would a critic say?”
• Ask for what it left out. “What’s the strongest argument against this recommendation? What have you not considered?”
• Ask it to stress-test the downside. “Walk me through how this goes wrong. What are the second-order consequences I’m not seeing?”
• Make it choose. If it gives you five options, ask it to rank them and justify the order, rather than hiding behind a neutral list.

Go back to the person weighing up a sales hire. Say the AI eventually recommends they hire. Before acting, they type: “Now argue the opposite. Make the strongest possible case that hiring is the wrong move right now.” The AI comes back with a sharper point than its own recommendation: that a new salesperson feeding leads into a weak conversion process just spends money faster without fixing the leak. They hadn’t asked that. The self-critique surfaced it.

This works because the AI isn’t attached to its first answer the way a person would be. It has no ego to bruise. Ask it to knock down its own reasoning and it will, often revealing the weak joint you’d have discovered three months later the hard way. One good self-critique is frequently worth more than the original answer.

You are still the one leading

One line to hold onto: you are the thought leader, and AI is the thought partner.

Everything above keeps a capable human firmly in the lead. The AI structures the thinking, asks the questions, drafts the answer, and even critiques its own work. You bring the judgment, the context, and the final call.

This matters in practice. When the AI makes the case against hiring, that’s a prompt to investigate, not an instruction to obey. You might know something it doesn’t, that your conversion is already strong and capacity genuinely is the ceiling. The AI raised the question and stress-tested both sides. You make the decision. It cannot carry your responsibility and it should never make the call for you. Used this way, it doesn’t replace your thinking. It sharpens it.

Try this before the next edition

You don’t need a budget or a strategy to start. Pick one decision or one problem you’re turning over right now, and run the full pattern:

• Give the AI a relevant role
• Tell it to interview you one question at a time, at least five questions, before it answers
• Answer honestly, in your own words
• When it responds, push back: ask it to critique its own answer and argue the opposite
• Treat what survives that as a strong first draft, not a verdict to accept

If you’re stuck on what to try it on, use this as your opening line: “I’m not sure how you can best help me this week. Interview me one question at a time, at least five questions, to work out where a thought partner would be most useful right now.” Let it find the problem with you.

The first time it asks you a question you hadn’t thought to ask yourself, and then dismantles its own advice when you push, you’ll understand why this is worth turning into a habit.

Over to you

I’m putting together a live AI session for surveyors, and I’d rather build it around what you actually want than guess.

There’s a quick poll on my feed. One vote tells me where to point it. If you’ve got a minute, add a comment too, with the one thing you’d want an hour on. The comments usually shape the session more than the poll does.

Vote here: https://www.linkedin.com/feed/update/urn:li:activity:7492189972888834048/