How to Prepare for a Job Interview with AI: Mock Interviews Step-by-Step (2026)

Two years ago I watched a friend walk into a job interview in Munich, get asked “tell me about a time you disagreed with your manager,” and freeze so completely that the interviewer offered him water. He was qualified. He was prepared — he’d read every listicle about interview tips. He just had never said his answers out loud to anyone.

That failure is why I’m writing this. Interview advice is everywhere; interview reps are rare. And that’s the one thing AI is genuinely brilliant at: being an infinitely patient sparring partner that asks you the hard questions at 11pm on a Tuesday, in your pyjamas, without judging.

I gave myself three evenings and three different AI interviewers to test which approach actually builds interview skill — not interview trivia, but the ability to think out loud under pressure. Here’s the method that came out of it.

Evening one: the text sparring round

I opened ChatGPT, pasted a real job description (a marketing role in Amsterdam — “3+ years experience, stakeholder management, data-driven mindset”), and used this prompt:

“You are a hiring manager at a Dutch scale-up, interviewing me for this role. Ask me one question at a time. After each answer, give me a score out of 10 and one specific thing to improve — then ask the next question. Start with ‘tell me about yourself.'”

The first question landed and I typed my answer. Then the score came back: 6/10. “Too long. You spent 40% of your answer on your university years. A hiring manager has already read your CV — start from your most recent role.”

It was right. And here’s what a listicle can’t give you: the sting of a 6/10 makes you fix it now, in the next answer. By question four, my answers were half the length and twice as specific. When we tried the same exercise a week earlier with a generic “give me interview tips” prompt, we’d gotten a pleasant list we immediately forgot. The scored, one-question-at-a-time format is the whole difference.

One detail worth stealing: ask the AI to play a specific interviewer, not “an interviewer.” “A skeptical German engineering manager who values precision” asks wildly better questions than “a hiring manager.” The persona changes the questions. Try both and you’ll feel it.

One caution from this round: the AI will happily accept vague answers. When I wrote “I’m a real team player,” it gave me 7/10 and moved on. Real interviewers don’t. Add this line to your prompt: “Push back on any answer that lacks a specific example. Ask ‘can you give me a concrete example?’ until I provide one.” That single sentence turned a polite chatbot into something resembling a real panel. (For the rest of the application — CVs, cover letters — our 40 ChatGPT prompts for job seekers pairs well with this method.)

Evening two: voice mode, where it gets uncomfortable

Text is safe. You can edit, pause, think. Real interviews don’t let you. So on evening two I switched to voice mode — ChatGPT’s voice conversation on my phone — and did the whole thing out loud, pacing my kitchen.

It was awful. Genuinely awful. Twice the AI cut me off mid-sentence because I paused to think — voice mode interprets a two-second silence as “your turn is over,” which is exactly what a bad interviewer does but not what good preparation should feel like. The fix: I learned to fill pauses with a steady “so, the key point here is…” — a trick that also works on humans.

But here’s what voice mode gave me that text never could: I heard my own filler. “Um,” “like,” “you know” — I counted eleven “ums” in a two-minute answer about stakeholder management. Eleven. No article about interview tips has ever made me count my ums. After three voice rounds, I was down to three, because nothing motivates like hearing yourself.

The other discovery: voice mode rewards the STAR method (Situation, Task, Action, Result) more than text does. When I rambled, the AI’s follow-up was always “and what was the result?” — because spoken rambling is more obvious than written rambling. If you only do one AI prep exercise before your next interview, make it one 20-minute voice round. It’s the closest thing to real pressure that doesn’t require a human.

Evening three: the hostile panel

For the final evening I asked Claude to be the interviewer from hell: “Ask me the questions candidates hate. ‘What’s your greatest weakness?’ ‘Why should we hire you over someone cheaper?’ ‘Why did you leave your last job?’ After my answers, tell me what a hiring manager would actually think, not what’s polite.”

The “greatest weakness” exchange was the most useful five minutes of the week. My answer — “I’m a perfectionist” — got dismantled. Claude’s verdict: “Every hiring manager has heard this 500 times. It signals you didn’t prepare, which is worse than any real weakness.” Then it helped me rebuild the answer around a genuine, minor weakness with evidence of management: “I used to over-commit to deadlines; now I add a 20% buffer and I’ve hit every deadline for 18 months.” Specific. Honest. Boring in the best way.

This is where I’ll plant my flag, because I know it’s contested: do not ask AI to write your answers. Every “perfect answer” the chatbot generates for you is a trap. Memorised answers sound memorised — interviewers can smell them in one sentence, and one canned answer poisons the credibility of everything else you say. Use AI to interrogate your thinking, to score your delivery, to find your ums and your waffle. Never to script you. The candidates who get hired aren’t the ones with the best answers; they’re the ones who sound like they’ve actually thought.

The whole method in thirty seconds

Pick a real job ad. Paste it. Tell the AI: one question at a time, score me, push back on vagueness.

Do it in text first. Then do it out loud — voice mode, kitchen, pacing.

End with the questions you dread. Rebuild the worst answer until it’s boring and true.

That’s it. Three evenings. No course, no coach, no €300 workshop.

Your prep kit: prompts and a scoring rubric

The setup prompt (works in ChatGPT, Claude, or Gemini):

“You are [specific interviewer persona] interviewing me for [paste job title + 3 key requirements]. Ask one question at a time. Score each answer out of 10 with one specific improvement. Push back on vague answers until I give a concrete example. Start now.”

The voice drill: same prompt, voice mode, 20 minutes, phone in hand, walking. Count your filler words after. Aim to halve them by round three.

The weakness rebuild: “Here’s my answer to ‘what’s your greatest weakness’: [paste]. Tell me what a hiring manager would actually think, then help me rewrite it around a real, minor weakness with evidence.”

Score yourself like this after each round:

• Clarity (did I answer the actual question?): /10
• Evidence (did I give a specific example?): /10
• Brevity (under 90 seconds per answer?): /10

If evidence is your lowest score three rounds running, that’s your project for the week — go mine your work history for stories before the next session.

One last thing, and it matters more than the prompts: do your final mock round the evening before the interview, not the morning of. A morning session leaves you second-guessing; an evening session lets your brain consolidate overnight. I can’t prove the neuroscience, but every candidate I’ve watched do this walks in calmer. Calm is half the interview. And before you walk in, make sure the interviewer likes what they see online — our guide to optimising your LinkedIn profile with AI covers that side of the prep.

Written by the PromptSkill Team — we test the tools we write about, so you get methods that survived contact with reality, not just theory.