What you’ll be able to do

Right, so by the end of this you’ll have run a proper mock interview with an AI, got feedback that actually holds up, and worked out a plan for the two questions that make your stomach drop. That’s it. No fluff, just a way to practise before the real thing happens.

Before you start

An AI assistant can’t tell you what your interviewer had for breakfast or whether they’re in a good mood that day. It doesn’t know your industry’s specific culture, and it definitely doesn’t know the actual humans on the other side of the table. What it’s genuinely good at is generating likely questions, listening (or reading) your answers, and telling you honestly whether you rambled or nailed it.

Think of it like practising a speech in front of a mirror, except the mirror talks back and occasionally tells you your STAR answer was 80% “S” and no “R”. That’s useful. That’s the whole point.

You’ll need your CV and the job advert to hand, that’s really it. No prior AI experience needed, promise.

Get set up

There isn’t one single best tool here, it depends what kind of practice you want.

If you want a spoken back-and-forth that feels closest to a real interview, ChatGPT’s voice mode or Gemini Live will do a spoken mock interview with you. You talk, it responds, out loud. That’s the closest thing to rehearsing the real conversation.

If you’d rather type it out, any of the big chat tools (ChatGPT, Claude, Gemini, Copilot) work fine for a text-based version, and honestly for a beginner that’s the easier starting point. Less pressure, more time to think.

Google also has a free tool called Interview Warmup, built specifically for this, no faffing about with prompts. And if you want something more structured and are happy to pay for it, Yoodli and Big Interview are dedicated practice platforms built around exactly this problem. I won’t quote prices, they change, just go check.

Before you start any of these, gather two things: your CV, and the actual job advert. Not a similar job, not “roughly the same role”, the actual advert for the actual job you’re going for. This matters more than people expect, because generic interview questions are everywhere online and mostly useless. The good stuff comes from pasting in specifics.

Try it yourself

Here’s where it gets practical. Three prompts, three different jobs, use them in order.

First, get your questions. Paste the job advert and a short version of your CV (bullet points are fine) and ask for the most likely questions.

I'm preparing for an interview for this role. Here's the job advert:
[paste job advert]

Here's a summary of my experience:
[paste CV or bullet points]

Give me the 10 most likely interview questions for this specific role, based on the responsibilities and requirements listed. Mix technical and behavioural questions.

Answer them out loud. Actually out loud, not in your head. This is the bit people skip and it’s the bit that matters most, because speaking an answer is completely different from thinking one. You’ll hear yourself waffle in a way you’d never notice silently.

Second, once you’ve answered a few, ask for feedback using STAR, that’s Situation, Task, Action, Result, the standard structure for behavioural answers.

Here's my answer to the question "[paste the question]":
[type out roughly what you said, or paste a transcript if you recorded it]

Give me honest feedback on this answer using the STAR method. Where was I too vague? Did I include a clear result? Be specific about what to cut and what to add.

Third, and this is the one that actually toughens you up: ask it to stop being nice.

Act as a sceptical interviewer for this role. I've just given an answer that sounds too polished or rehearsed. Push back with a follow-up question that a real interviewer might ask to test whether I actually did what I claimed.

That third prompt is the difference between practice that feels good and practice that works. Anyone can get an AI to say “great answer, well structured!” What you want is something that pokes holes, the way a real interviewer might when your answer sounds a bit too smooth.

Do this for the two questions you dread most. You know which two. Everyone has them, the “tell me about a time you failed” one, or the “why did you leave your last job” one that requires some careful phrasing. Rehearse those specifically, more than once.

Check the result

So how do you know if the feedback you’re getting is any good, rather than just generic padding?

Test one: ask it to be specific. If the feedback says something like “be more concise” or “add more detail”, push back and ask exactly where. A useful response points at the actual sentence. A useless one just repeats the same vague advice in different words.

You said my answer was too vague. Point to the exact sentence where I was vague, and rewrite it as an example of what a stronger version would sound like.

Test two: ask if the questions it gave you are actually role-specific or just generic interview questions with the job title bolted on. You can literally ask this.

Are these questions genuinely tailored to this specific job advert, or are they generic interview questions? Be honest, and if any are generic, replace them with ones based on the actual responsibilities I pasted.

You’ll get an honest answer more often than you’d think. If it says “actually, questions 3 and 7 are fairly generic”, that’s useful, that’s the AI doing its job properly.

Test three, and the real one: record yourself answering out loud, listen back, and see if you’d believe you. If you sound rehearsed and robotic, that’s worth fixing before feedback from any tool. No amount of AI polish helps if the delivery sounds like you’re reading off a script.

You’ll know you’re ready to move on when you can answer your two dreaded questions without the pause-and-panic feeling, and when the sceptical-interviewer follow-ups stop catching you out.

If it doesn’t work

A few honest warnings here.

Don’t paste anything with sensitive personal information you wouldn’t want stored somewhere, salary details from a current contract, references’ personal contact details, that sort of thing. Your CV and a job advert are fine, they’re not secrets. But if you’re pasting internal company information from your current job to compare roles, or anything under an NDA, don’t.

Keep your real application separate from your practice sessions. It’s easy to end up editing your actual CV inside a chat window because it’s convenient, then losing track of which version is the one you sent off. Do your practice in one place, keep your submitted documents somewhere else, properly saved.

If the questions feel too generic despite pasting a detailed job advert, the fix is usually more detail, not a different tool. Add company context if you have it, add the actual seniority level, add specifics about the team. Vague input gets vague output, every time.

If the feedback misses the mark, or feels like it’s just agreeing with everything you say, that third prompt (the sceptical interviewer one) is your fix. Ask it directly: “stop being encouraging, tell me what’s actually wrong with this answer.” Most tools will do this if you ask plainly enough.

And the big one: an AI does not know who’s interviewing you. It can’t predict the actual questions, the actual mood in the room, or the actual person deciding whether you get the job. Use this for practice, structure, and confidence, not as a crystal ball. If you walk in expecting exactly the ten questions you rehearsed, you’ll be thrown by the first one that isn’t on the list. Practise the skill, not the script.

Keep exploring

If you haven’t already sorted your CV, there’s a separate guide on this site called How to get AI to write my CV, worth doing before this one really. There’s also How to get AI to find and apply for jobs, which covers the bit before the interview even happens. And if you want a broader look at getting AI to actually tell you the truth rather than just being agreeable, How to get AI to be honest with me covers that in more depth than I could fit here.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.