The default AI mock interview is useless: you say something vague, the model replies “great answer”, and you walk into the real loop convinced you’re ready. Then a real interviewer probes once and you fold. These prompts force adversarial mode — the model has to push back on capacity estimates, refuse “increase engagement” as a metric, name the failure mode you didn’t mention, and at the end produce a scored rubric you can act on. Pair them with the interview debrief prompts so each mock becomes a drill list.
TL;DR
- Paste the prompt that matches your round, replace the bracketed placeholders, and tell the model your level (junior / senior / staff) — without it, models default to easy questions.
- Run it in ChatGPT Advanced Voice Mode for verbal rounds: ~2-3 second response latency makes it close to a real call (as of June 2026). Free gets a short daily preview, Plus ($20/mo) gets several hours a day, Pro gets near-unlimited.
- For take-home and system-design rounds, paste your resume + the job description first so the mock is grounded — Claude Opus 4.7, Sonnet 4.6 and Gemini 3.1 Pro all hold a 1M-token context, so the whole packet fits.
- Always demand a scored rubric at the end. The score is the point: it converts the mock into a list of things to drill.
Best for
- SWE / PM / designer interview prep
- Career switchers entering unfamiliar interview formats
- Senior-to-staff IC promotion interviews
- Internship-to-full-time conversion loops
- Take-home submission review before sending
Which AI to use for which round
The prompts below are model-agnostic, but the round changes which tool wins (June 2026):
| Round | Best tool | Why |
|---|---|---|
| Behavioral / verbal | ChatGPT Advanced Voice Mode | ~2-3s latency, interruptible; closest to a live call |
| System design | Claude Opus 4.7 or Gemini 3.1 Pro | 1M context fits your resume + JD + a design doc; strong on trade-off pushback |
| Take-home code review | Claude Sonnet 4.6 or Opus 4.7 | Reads the full repo paste, flags edge cases, scores README + tests |
| Product sense (PM) | Any frontier model in voice | Conversational metric pushback works well spoken |
Voice quality matters here: ChatGPT’s Advanced Voice Mode is the most natural for back-and-forth, Gemini Live is a solid second for spoken roleplay, and Claude’s voice mode is improving but still laggier (multi-language and push-to-talk are rolling out through 2026). For text-only deep rounds, lead with Claude or Gemini for the longer context.
Tip: start the session by pasting your real resume and the job posting, then add “Ground every question in this material.” A grounded mock surfaces the gaps a generic one never will.
1. Behavioral mock with follow-ups
You are a senior {role} hiring manager at {company tier}. Run a 30-minute behavioral mock. Ask me 4 questions one at a time. After each of my answers, push back with 2 follow-up probes. Do not be polite — be rigorously curious. At the end, score me on STAR completeness, signal density, and red flags.2. System design mock
You are a {company} staff engineer running a 45-minute system design mock. Topic: "{system}". Step 1: ask me to clarify scope (do not give me requirements; make me ask). Step 2: push me on capacity estimation. Step 3: drive me to specify trade-offs in 2 of my choices. End with a scored rubric.Score it against the way real 2026 rubrics weight the round: requirements gathering ~15%, high-level architecture ~25%, and the deep dive ~30% (the single highest-weighted dimension). At the staff bar, a “strong hire” drives the conversation and names failure scenarios before the interviewer asks — tell the model to dock you when you wait to be prompted.
3. PM-product-sense mock
You are a {company} PM interviewer. Ask me a product-sense question: "How would you improve {product}?" Push me on: (a) target user, (b) chosen metric, (c) why this beats alternative features. Don't accept "increase engagement" as a metric.4. Role-specific deep-tech mock
You are a senior {specialty — e.g., ML platform engineer} interviewer. Run a 30-min mock with 3 questions calibrated to senior level. After each, probe for the failure mode I did not mention. Do not lead. End with a rubric: technical depth, blind spots, communication.5. Behavioral red-flag-finder mock
Be a skeptical interviewer. I will give you a STAR answer. Find the 3 red flags a real hiring manager would notice (vague metrics, no ownership, no learning, blame-shifting). Be specific about which sentence reveals each flag.
Answer: "{paste}"6. Cold-call adversarial mock
Ask me a curveball question I would not prepare for in a {role} interview at {company}. Don't tell me what category it is. After I answer, tell me which interview signal that question tests, and how my answer scored.7. Role-fit / motivation mock
You are a hiring manager focused on motivation and role-fit. Ask me 5 questions designed to surface whether I actually want this {role} or am applying out of convenience. Then write 100 words on whether you would advance me, and why.8. Pair-programming mock
You are a senior engineer running a 45-min pair-programming mock. Task: "{task}". Ask me to write code step-by-step. After each chunk, run it mentally and find 1 bug or edge case I missed. End with a hire/no-hire writeup explaining the technical signal you observed.9. Take-home review mock
I just finished a take-home assignment for {role}. The brief was: {brief}. My submission: {paste}. Pretend you are reviewing it for a real hiring committee. Score on: correctness, code quality, testing, README, scope discipline. List the 3 questions you would ask me in the follow-up.10. Late-stage hiring-manager round
Play a hiring manager doing the final round before offer. Ask me 4 questions designed to test: strategic thinking, scope-of-ownership, how I handle a failed project, what I will need from them. Push back if I am too rehearsed.11. Senior-to-staff promotion mock
Run a senior-to-staff IC promotion mock. Ask me 3 scope questions where the right answer requires going wider than my immediate team. Then assess my answers against a staff bar: cross-team influence, strategic decisions, written artifacts. Brutally honest scoring.12. Reverse-mock: I interview AI
Swap roles. I will interview you. You are applying for {role} at {company}. I ask, you answer. After each round, tell me whether my question was sharp or weak, and what a better question would have been.13. Post-mock written feedback
Below is the transcript of a mock interview we just did. Write a 300-word post-mock feedback report. Sections: (1) strongest answer + why, (2) weakest answer + what was missing, (3) top 3 patterns I should drill before the real interview, (4) one habit to drop.
{paste transcript}Common mistakes
- Letting AI play “nice interviewer” so every vague answer comes back as “great point”
- Not requesting follow-up probes — real interviewers push twice, your mock should too
- No scoring rubric at the end, so the practice gives you no drill list
- Practicing only behavioral and skipping system design / product sense
- Using the same mock prompt for every role and company tier
- Not specifying a level (junior vs senior vs staff) — the model defaults to easy
- Skipping the grounding step, so the model invents a generic candidate instead of probing your real resume
FAQ
Which model should I use for a verbal mock? ChatGPT in Advanced Voice Mode is the most natural for spoken back-and-forth, with roughly 2-3 second response latency and the ability to interrupt (as of June 2026). Free users get a short daily preview, Plus ($20/mo) gets several hours a day, and Pro ($200/mo) is effectively unlimited. Gemini Live is a strong alternative for voice roleplay.
Can the AI really push back like a tough interviewer? Only if you tell it to. Frontier models default to agreeable. The prompts here explicitly instruct it to find red flags, refuse weak metrics, and dock you for unprompted gaps — that single instruction is what turns “great answer” into useful pressure.
How do I make the mock specific to my role instead of generic? Paste your resume and the exact job description at the start of the chat, then add “Ground every question in this material.” With 1M-token context on Claude Opus 4.7, Sonnet 4.6 and Gemini 3.1 Pro, the whole packet plus a design doc fits in one session.
Is a free plan enough for this? For text mocks, yes — ChatGPT Free (GPT-5.5) and Claude Free run every prompt here. The paid tiers mainly buy more Advanced Voice minutes and longer in-app context, which matter most for long verbal loops and document-heavy take-home reviews.
Do these replace a real mock with a person? No. Use them to drill the patterns the AI flags, then validate delivery with a human (a peer, a mentor, or a paid platform). The AI’s job is volume and an honest score; a person catches the human cues a model still misses.
Related
- Behavioral question prompts
- STAR interview prompts
- Tell-me-about-yourself prompts
- Interview debrief prompts
- Self intro prompts
- Interview Question Prediction Prompts: 12 Templates to Prep Smarter
- Why This Company Answer Prompts: 12 Templates Beyond “I Admire Your Mission”
- AI Mock Interview: Realistic Practice, Follow-Ups, Scored Feedback