Practising for interviews with an AI: what works, what doesn't

There are now half a dozen AI mock-interview tools. Some are good. Most are gimmicks. The difference is mostly how you use them.

We just shipped voice mock interviews on CVCL — pick a CV + paste a JD, talk to a tailored AI hiring manager for 5-20 minutes, get a written debrief afterward, listen back to the recording. Building it taught us a few things about what actually moves the needle when you're prepping for a real interview.

Why voice beats text-based "interview prep"

Text-based prep — reading practice questions, writing out STAR stories, even chatting with ChatGPT in text — is fine for knowing the material. It does almost nothing for the interview itself, which is a real-time spoken conversation under pressure.

Voice practice forces three things text prep can't:

1. Verbal latency. You have to start your answer in 2-3 seconds, not 30. If your STAR stories aren't memorised at the level of "I can start speaking before I've finished thinking," you'll fumble in the real thing.
2. Speaking clearly under pressure. Every recruiter we've talked to says half their no-hires aren't due to bad answers — they're due to good answers delivered in a way that sounds nervous, hedging, or rambling. You can't fix that in text.
3. Real follow-up dynamics. A real interviewer will interrupt your STAR story mid-sentence with "wait, what was the actual metric there?" Practising with a bot that does the same thing builds the muscle to handle it without freezing.

If you've only ever rehearsed silently in your head or in writing, you'll hit a wall the first time you say your story out loud and realise you can't get through it cleanly.

The mock-interview playbook that actually works

1. Tailor the session to a SPECIFIC job, not "generic backend interviewer"

Generic mock interviews give generic preparation. Take a real JD you're applying to (or a realistic stretch one). Paste it. Let the bot ask questions drawn from THAT role's requirements.

Why this matters: a Staff Backend role at a payments fintech and a Senior Backend role at a consumer ML startup will probe completely different things. The keywords, the tradeoffs, the depth of system design — all different. Practising against one is barely practice for the other.

2. Pick a persona that matches the company you're targeting

Different companies interview in different styles:

• Google / FAANG-style — bar-raiser format, structured behaviourals, deep system design. Lots of "tell me about a time when..." with sharp follow-ups on scope and impact.
• Startup — scrappy, hypothetical-heavy, tests for ownership and resourcefulness. "How would you build X from scratch in two weeks?"
• Consulting (MBB) — structured case-interview format, hypothesis-led, MECE, push-for-recommendation.
• Tough — generic but unforgiving. Useful as a stress test before any high-stakes interview.
• Friendly — eases you in. Good for the first practice run when you haven't said your stories out loud in a while.

If you're interviewing at Google next week, don't practise with the friendly persona. The real interview will be much sharper.

3. Use SHORT sessions to drill, LONGER ones to build endurance

A common mistake is thinking the only valuable practice is a full 30-45 min mock interview. Often the opposite is true:

• 5-minute sessions — drill specific things. Your elevator pitch. Your "why do you want to leave your current role" answer. The one weak STAR story you keep stumbling on.
• 10-minute sessions — covers the opener + one deep technical probe. Useful when you want to practise a specific question type without committing 30 min.
• 15-20 minute sessions — full simulation. Use these the day before a real interview to build endurance and check timing.

It's much more efficient to do six 5-min drills on different weak spots than one 30-min session where you mostly get reps on things you already do well.

4. After every session, listen to the recording

This is the single highest-value practice habit and it's also the most uncomfortable.

When you listen back you'll hear things you can't notice in real time:

• "Um" / "uh" / "like" / "basically" — count them, you'll be horrified
• Long sentences with no breath, where you should have stopped at a period
• Hedging language that makes you sound less senior ("I think maybe we sort of...")
• Fast nervous talking when the right move is to slow down and be deliberate

Most candidates have never heard themselves answer interview questions out loud. Doing it once is a wake-up call.

5. Read the written debrief after, not during

A good AI mock interviewer should give you written feedback after the session — what answers landed, what answers didn't, model rewrites for the weak ones. Read it carefully BEFORE your next session. Don't just queue up another one immediately.

The point is to fix specific things, not log mock-interview hours.

6. Practise the salary negotiation separately

The dynamics of a salary negotiation are completely different from an interview. The interviewer is now a hiring manager who has decided to make you an offer; their job is to anchor low and see if you push back.

Most candidates have never practised this out loud. The first time you say "I appreciate the offer, but based on the market data for senior backend engineers in this region, I was looking for X" — you want it to be the SECOND time you've said it, not the first.

CVCL has a salary-negotiation persona for exactly this. Most other tools don't.

Failure modes to avoid

Mistake 1: Treating it like a video game

Mock-interview tools have a clear scoreboard (the feedback grade). It's tempting to optimise for the score. Don't. The point is to prep for a real interview, not to get a 9/10 from the AI. A 6/10 honest practice run with a hard persona is more useful than a 9/10 with the friendly one.

Mistake 2: Practising the same easy questions over and over

The behavioural questions you already nail aren't the ones that'll trip you up in the real interview. Spend most of your practice time on the ones that scare you — the gap question about the skill the JD wants but you don't have, the "tell me about a failure" you don't have a clean story for.

Mistake 3: Skipping the recording

If you don't listen to your recordings you lose 80% of the value. Schedule 10 minutes after each session specifically for playback.

Mistake 4: Doing one massive 60-min session the night before

Doesn't work. Spread practice over 5-7 days. The compounding effect of multiple short sessions is much stronger than one marathon. (This is true for any practice — sports, music, language learning, public speaking.)

Try it

CVCL's voice mock interview is at cvcl.online/interview-prep. Pick a CV, paste a JD, choose persona + duration + voice, hit start. Costs scale with session length (1 credit per minute — a 5-min drill is 5 credits, a 20-min full sim is 20). Sessions are recorded; you can play back from the History page.

If you've never done a voice mock interview before, start with a 5-minute session in friendly mode. Then make it harder.

TL;DR

• Voice practice trains things text practice cannot — verbal latency, clear delivery, real-time follow-up handling
• Tailor each session to a specific JD + match the persona to the company style
• Short focused drills > long marathon sessions
• Always listen back to the recording (this is where the real learning happens)
• Practise salary negotiation SEPARATELY, before you have an actual offer
• Spread practice over a week, not the night before