Practical

How to use AI to prepare for a job interview

By the Keep Your Head desk · About a 6-minute read

Almost everyone uses AI the same way here: "what questions will they ask for a [job title] role?" You get a tidy list, you read it, you feel prepared, and then in the actual room you're assembling an answer from scratch anyway. Reading is not rehearsing. The tool is capable of something much better — being a difficult interviewer at 11pm, for free, as many times as you can stand.

Why the question list is the weakest use

Two reasons. First, you already know most of those questions — tell me about yourself, biggest weakness, why us. Reading them again isn't new information.

Second and more importantly, interviews aren't lost on the first question. They're lost on the follow-up — the "what was your specific contribution?" after you said "we improved retention," the "what would you do differently?" after a story that was going fine. That's where a rehearsed answer runs out and you start improvising. A list of questions never touches this. A conversation does.

Round one: make it interview you

One question at a time, and make it wait. That constraint does most of the work — otherwise the model dumps ten questions and answers them itself, which is the failure mode you're trying to escape.

You're interviewing me for this role. Job description
below.

Rules:
- Ask ONE question, then stop and wait for my answer.
- After each answer, ask one follow-up that probes the
  weakest part of what I said.
- Do not give me feedback yet and do not write answers
  for me.
- Go for 8 questions, getting harder.

Job description: [paste it]
My background: [3-4 lines]

Answer out loud, typing roughly what you'd actually say — not a polished paragraph. The point is to find where you stall, and you can't find that by writing.

Round two: ask what was weak

Now switch it to assessor. This is where the real value is, because it will name things you can't see about your own answers:

Now assess that interview honestly. For each answer:
- What was the strongest single sentence?
- What would a sceptical hiring manager doubt?
- Where did I describe a situation but not my own
  specific contribution?

Then: which two answers most need rewriting, and why?
Be blunt. Don't reassure me.

"Where did I describe a situation but not my own contribution" catches the most common interview weakness there is — the candidate who narrates a team's success and never says what they personally did. That single question is worth the whole exercise.

Note the "don't reassure me." Left to its defaults, a chatbot grades generously, because it's tuned to be agreeable. You have to ask for friction explicitly; it's the same design pull described in using AI as a therapist.

Round three: the questions you ask them

Most candidates ask something safe about culture. You can do better by working out what you actually need to know before accepting this job:

Based on this job description, what are the three things
about this role that would most likely make it a bad fit
for someone — things the posting doesn't say?

Turn each into a question I can ask that sounds curious
rather than suspicious.

Read a job posting closely and it tells you where the pain is: a role that's been open a long time, an oddly broad remit, "wears many hats," a manager title with no team. Questions built on that are more memorable than anything about culture, and they're genuinely for you — the interview goes both ways, and a bad fit costs you a year.

Where it will mislead you

It doesn't know this company. It knows the shape of companies like this. Anything specific about their strategy, their culture, or their interview process is a plausible guess. Verify from the actual site, recent news, and people who work there.
Don't memorise its phrasing. Rehearsed AI wording sounds like rehearsed AI wording — the same problem as in cover letters. Keep your structure and your own words. The tool's job was to find the gaps, not to supply the sentences.
Salary figures are unreliable. Ranges vary by location, year, and company size, and the model will produce a confident number anyway. Use real sources for this one.
Strip the confidential bits. If your best story involves a client or unreleased work, anonymise it before it goes in the box — the 10-second rule.

Keep your head:

Keep your head

Don't ask for a list of questions — make it ask you, one at a time, and follow up on your weak spots. Then ask what was weak, and insist it be blunt. You want the uncomfortable version of that conversation to happen in your kitchen, not in the room.

Get one of these a week. Our free newsletter sends one genuinely useful AI habit and one judgment check every week — no hype, four-minute read. Subscribe on the home page.