TL;DR: Use a general chatbot’s voice mode as your interviewer, not a scripted “interview” app. ChatGPT Plus ($20/mo) or Claude Pro ($20/mo) both run ~45-minute spoken mocks and hold a transcript; Google’s free Interview Warmup was retired in April 2026, so the free route is now Gemini Live or a daily voice preview. The prompt below forces the model to press on vague claims, score five dimensions, and name the one sentence to rewrite. The single biggest mistake is asking for a “friendly” interviewer — friendly nods you through and you leave underprepared.
The task
Your behavioral round at the company you actually want is on Thursday. You’ve rehearsed the “tell me about yourself” opening four times alone in the shower. You don’t have a friend with the time to do a real mock, and the one friend who’d be good already burned an hour on you last week. You need a 45-minute mock interview that asks five behavioral questions, presses on the vague parts of your answers, doesn’t let “I worked closely with the team” slide, and ends with scored feedback that names the exact sentence you need to rewrite, not “be more concrete.”
Where AI helps — and where it does not
A general-purpose chatbot in voice mode sustains a mock-interview cadence for 30-60 minutes, asks realistic follow-ups, presses on vague claims (“you said ‘I led’ — were you the EM, the TL, the sole IC? Who reported to you on this?”), and scores along dimensions you specify. It also switches interviewer style on command: terse Amazon LP-style, conversational consulting-case-style, time-pressured skeptic. What AI cannot do is replicate the company-specific signals that come from real interviewers. Without 2-3 real interview reports (Glassdoor, Levels.fyi, peer notes), the mock pulls toward a generic FAANG-shaped round that may not match your target.
The named failure mode: the polite interviewer. AI nods through weak answers (“Great example! Next question…”) instead of pressing. You leave the mock confident and underprepared. Force the prompt to press on at least 2 vague claims per answer and to deduct points when the follow-up isn’t answered.
Pick the right tool (June 2026)
Skip the dedicated “interview” apps unless you want their analytics layer — the live mock itself is better in a general chatbot’s voice mode, because you control the rubric and it holds the whole transcript.
| Tool | Cost (June 2026) | Voice mock | Holds transcript | Best for |
|---|---|---|---|---|
| ChatGPT Plus (GPT-5.5) | $20/mo | Advanced Voice, long sessions | Yes | Most behavioral + PM/case mocks |
| Claude Pro (Sonnet 4.6 / Opus 4.7) | $20/mo ($17 annual) | Voice on iOS/Android/web | Yes | Long answers, structured rubric grading |
| Gemini (3.1 Pro) | AI Pro $19.99/mo | Gemini Live, hands-free | Yes | Paste a full casebook (1M-token context) |
| ChatGPT Free | $0 | ~15-min daily voice preview | Yes (text) | One short mock, then switch to text |
| Yoodli | Free (5 roleplays) / Pro ~$24/mo | Yes | Yes | Delivery analytics (filler words, pace) |
Notes, as of June 2026: ChatGPT Free’s Advanced Voice runs on a short daily preview (~15 minutes), so it’s fine for one quick round but not a full 45-minute mock. Google’s Interview Warmup was retired in April 2026; the free conversational replacement is Gemini Live (Career Dreamer covers exploration). Gemini 3.1 Pro’s 1M-token context is the standout for case prep: paste an entire casebook and have it quiz you case by case. Yoodli grades how you speak (pace, filler words, eye contact via webcam), not what you say — pair it with a content-graded chatbot mock, don’t substitute it.
What to feed the AI
- Role and level you’re interviewing for (Senior PM, Staff Engineer, Strategy Consultant)
- Company type and 1-2 specific cultural signals (Amazon LPs, Stripe writing-test, McKinsey case style)
- 3 themes to practice (conflict, ambiguity, executive influence, technical depth)
- Your 3 weakest answer areas (where you tend to drift, hedge, or run long)
- The interviewer persona (friendly, terse, skeptical, time-pressured)
- Mock format (voice for closer-to-reality, or text for a transcript)
- 2-3 real interview reports for that company, even rough notes
- Scoring rubric you want (structure / content / delivery / impact / ownership)
Copy-ready prompt
You are conducting a behavioral mock interview for a [role] at a [company type], calibrated to this company's actual interview style (cultural signals below).
Company cultural signals: [paste 2-3 sentences from real reports — what they grade on, how they push back]
Role + level: [Senior PM, Staff Engineer, etc.]
Themes to practice: [conflict, ambiguity, executive influence, technical depth — pick 3]
My weakest areas (press here): [3 areas where I drift, hedge, or run long]
Interviewer persona: [terse / skeptical / time-pressured — never "friendly"]
Mock format: [voice / text only]
Rules of engagement:
1) Ask 5 behavioral questions, one at a time. Wait for my full answer before moving on.
2) After every answer, ask 1-2 realistic follow-ups before the next question. At least 1 follow-up per answer must press on a vague claim ("you said 'I led' — what does that mean here?", "what would the engineer you replaced say about this?").
3) Do not nod through weak answers. If I hedge, hand-wave, or run past 2 minutes, interrupt politely with the follow-up.
4) Track which themes you've covered. Hit all 3 of my requested themes across the 5 questions.
5) Final feedback (after question 5) must include:
- Strongest answer + the specific sentence that landed
- Weakest answer + the specific sentence to rewrite + a 1-sentence rewrite suggestion
- Scores 1-5 on: structure / content / delivery / impact-claimed / ownership-language
- 3 follow-up questions I should expect from a real interviewer on my weakest answer
- 1 "rehearse this sentence aloud 5 times before Thursday" line
Tone: this should feel uncomfortable in 1-2 places. That's where I grow.Voice tip: in ChatGPT Advanced Voice or Gemini Live, paste this prompt as text first, then switch to voice for the actual mock — the model keeps the rules and you get a clean transcript to study after.
Shorter variant — single-question deep press
Conduct a deep press on one behavioral answer.
Question to ask first: [paste the question]
Role + level: [paste]
After my answer, ask 4 follow-ups in sequence, each pressing on the weakest part of my prior answer. Then score 1-5 on structure / content / ownership. End with the one sentence I should rewrite.Sample output
A useful follow-up: “You said ‘we shipped on time.’ Who decided to cut the analytics tier from v1 — you, your manager, or the team in a group call? And what would your manager say was the riskiest cut in retrospect?”
A useful interruption: “Pausing you — that’s the third ‘collaborative’ I’ve heard in 30 seconds. Pick the one collaboration moment that was hardest and walk me through 60 seconds of what you did differently than the obvious play.”
A useful final feedback paragraph: “Strongest: Q3, the migration story. The sentence ‘I owned the rollback plan that we ended up not needing’ landed because it shows judgment without bragging. Weakest: Q1, the conflict story. The sentence ‘we eventually came to alignment’ is the rewrite target — passive, no agency, no specifics on what changed your peer’s mind. Suggested rewrite: ‘I asked her what would make her change her mind, she named two things, I delivered one of them in 48 hours, and she dropped the objection.’ Scores: structure 4 / content 3 / delivery 4 / impact-claimed 3 / ownership-language 2. Follow-ups to expect Thursday: who pushed back; what you’d do differently; whether the peer agreed at the time or only later. Rehearse this aloud 5x: ‘I asked her what would change her mind, she named two things, I delivered one in 48 hours.’”
How to refine
- Demand pressing follow-ups: “Press harder. Every vague claim (‘led’, ‘worked closely’, ‘collaborated’, ‘drove’) must trigger a follow-up before the next question. If I hedge twice in one answer, interrupt me.”
- Tune to company culture: “Recalibrate. This is a
[company]interview, not a generic one. Their bar is[one cultural signal — Amazon ownership, Stripe writing, McKinsey structure]. Re-grade my last answer against that bar.” - Score on what matters: “Add ‘ownership language’ to the scoring rubric. Deduct points when I say ‘we’ for something I clearly did alone, and when I say ‘I’ for something the team did.”
- Replay the weakest answer: “After feedback, give me the weakest question again. Score the second attempt and tell me whether it’s interview-ready or needs a third pass.”
- Run a hostile variant: “Re-run the same 5 questions as a skeptical interviewer who’s running 10 minutes late and wants the punchline first. Different style, same rubric.”
Common mistakes
- Asking AI to “be a friendly interviewer”. Friendly produces nodding, not pressing; tell it to interrupt politely on vague claims
- Skipping the cultural-signals input. Without it, every mock feels like a generic FAANG round and doesn’t prep you for Stripe, McKinsey, or your actual company
- Not getting written feedback at the end. The mock is half the value; the feedback is the other half
- Doing only text mocks. Voice exposes hedging, ums, and run-on sentences that text hides; use ChatGPT Advanced Voice or Gemini Live
- Trying to do a 45-minute mock on ChatGPT Free. The Advanced Voice preview caps at roughly 15 minutes a day; use it for one short round, then move to Plus, Claude Pro, or text
- One round, no iteration. The second attempt at the weakest answer is where the real improvement happens
- Letting AI invent your stories. Your stories must be real; AI should press them, not write them
- 6+ mocks before the real interview. Diminishing returns past 3; the rest is anxiety performance, not preparation
- Ignoring AI’s “this is the sentence to rewrite” line. That sentence is exactly the one a real interviewer will press on
FAQ
- Which tool should I use?: For a full voice mock, ChatGPT Plus (GPT-5.5) or Claude Pro (Sonnet 4.6), both $20/mo, hold long sessions and a transcript. Free route: Gemini Live, or ChatGPT Free’s ~15-minute daily Advanced Voice preview for one short round. Google’s free Interview Warmup was retired in April 2026, so that link no longer works.
- How long should a mock be?: 30-45 minutes for a behavioral round; 60 minutes for a technical screen; 90 minutes for a system design or product-sense round. Match the real interview length so your pacing instinct calibrates.
- Voice or text?: Voice is closer to reality and exposes filler words; text gives you a transcript to study. Best practice: voice once for realism, then read back the auto-transcript both ChatGPT and Gemini Live save. If you want delivery analytics (filler-word count, pace), run one round in Yoodli on top.
- How many mocks before the real interview?: 2-3 high-quality mocks with iteration on the weakest answer. More than 5 is performance anxiety, not preparation.
- What if my role is niche (research scientist, founder, late-career exec)?: Feed AI 2-3 real interview reports for that exact role. Without them, AI defaults to junior-mid IC patterns and misses the senior-specific bar (vision, organizational influence, strategic judgment).
- Should I tell AI my real story details?: Yes. Accuracy of the mock depends on real specifics. AI can press your story honestly only if it knows what actually happened. If privacy is a concern, anonymize names and metrics but keep the structure intact.
- Is paying for a $148/mo interview app worth it?: Not for practice. Tools like Final Round AI add real-time in-interview overlays, but for preparation a $20/mo general chatbot does the mock better because you set the rubric. Spend the difference on a coach for the rounds that matter.
Related
- Behavioral interview prep AI — behavioral specifics
- STAR interview answers — the underlying stories
- STAR follow-up questions — anticipate follow-ups
- Self introduction — opening 60 seconds
- Salary negotiation prep — once you pass the interview
- JD matching analysis AI — calibrate prep to the JD
- Tell Me About Yourself AI