AI mock interview practice: tools, a routine, prompts and a scoring rubric

What is an AI mock interview?

An AI mock interview is a practice session where a language model plays the interviewer: it asks questions for a specific role, listens to or reads your answers, pushes with follow-ups and gives feedback afterwards. The value is in saying answers out loud under mild pressure, which reading tips never gives you.

It's not the same as an interview copilot, which suggests answers during a real interview. A mock interview makes you better; a copilot answers for you, and many employers prohibit that (we cover the rules in are AI interview assistants allowed?).

Which AI mock interview tools work in 2026?

Start with a free voice model, then pay for structure only if you need it. Prices are from each tool's own page, checked October 2026.

That last row surprised us. Interview Warmup was the default free recommendation for years, including in our own earlier write-ups. When we loaded grow.google/interview-warmup this week, it returned a 301 redirect to a Grow with Google article on interview tips. Google's own Coursera course, "Prepare and Practice for Interviews with AI", now teaches mock interviews with Gemini Live, so that's the closest replacement from the same company. If a guide still sends you to Warmup, it hasn't been updated.

Which to pick? If you freeze or ramble, start with Yoodli, because it measures delivery. If you need many interview types quickly, Final Round AI's scenarios save setup. If you want to be taught a method first, Big Interview's course does that. For everything else, a voice model with good prompts covers more than people expect. Our roundup of the best AI interview prep tools compares more options, including peer mocks on Exponent and paid sessions on interviewing.io.

How do you run an AI mock interview, step by step?

This is the routine we'd give a friend with an interview next week. A full pass takes about 45 minutes.

1. Write the brief. Paste the job description, a short resume summary, the round type (recruiter screen, hiring manager, technical) and who the interviewer probably is. What goes wrong: skip this and the AI asks textbook questions you'd never get.
2. Set the rules before the first question. One question at a time, wait for your full answer, one follow-up per answer, no tips during the session. What goes wrong: without rules, most models answer their own question or coach you mid-answer.
3. Answer out loud, timed. Aim for one to two minutes on behavioral questions. Use voice mode or push-to-talk. What goes wrong: typing answers trains you to write, and nobody types in an interview.
4. Score at the end, against a rubric. Ask for scores per criterion with a quote from your answer as evidence. What goes wrong: feedback after every answer breaks the flow, and unanchored feedback is nearly always too kind.
5. Re-drill the two weakest answers the next day. Same question, new attempt, compare scores. What goes wrong: chasing new questions instead of fixing the answers that would actually lose you the offer.

Do this three times across a week and add one mock with a real person, even a friend reading questions off a list. AI is weaker at the strange follow-up a tired hiring manager asks at minute 40.

What prompts should you use for an AI mock interview?

These four cover a full session. Paste them into ChatGPT, Claude or Gemini, typed or spoken. For a bigger library, see our 36 ChatGPT prompts for interview prep.

Prompt 1, the interviewer setup:

You are the [hiring manager / recruiter / senior engineer] interviewing me for [job title] at [company].
Job description: [paste]
My background: [5-line resume summary]
Run a [30]-minute [behavioral / technical / case] interview. Rules:
- Ask one question at a time and wait for my full answer.
- After each answer, ask exactly one follow-up that probes a weak or vague part.
- Do not give tips, praise or feedback until I say "end interview".
Start with your first question.

Prompt 2, when it's being too gentle:

Be a tougher interviewer for the rest of this session. If my answer lacks a specific example, a number, or what I personally did, interrupt and ask for it. Ask "why" at least once per answer.

Prompt 3, scoring at the end:

End interview. Score each of my answers from 1 to 5 on: answered the question, structure, specifics, my own role, length, and follow-up depth.
For every score, quote the part of my answer that justifies it. Be strict: a 5 means a hiring manager would remember the answer.
Then list my two weakest answers and what a stronger version would include. Don't write the answer for me.

Prompt 4, the next-day re-drill:

Yesterday I scored low on this question: [question]. My answer was: [paste or summarize].
Ask me the same question again, then one follow-up. Score the new answer with the same rubric and tell me what changed.

Note the last line of prompt 3. If the model writes the "ideal" answer, you'll memorize its words and sound like a script. Ask what a better answer would include, then build it from your own experience.

How should you score mock interview answers?

Score every answer on the same six criteria from 1 to 5, using written definitions so the scores mean the same thing from one session to the next. Here's the rubric we use; it's the one prompt 3 refers to.

Our rule of thumb: once your core stories average 4 or better across two sessions, stop adding new questions and rest. Past that point, polishing tends to make answers sound rehearsed.

What do people get wrong when practicing with AI?

• Accepting the praise. Models default to encouragement. Without a rubric and quoted evidence, "great answer!" means nothing.
• Practicing in writing. A typed answer that reads well can take four minutes to say.
• Memorizing AI-written answers. Interviewers hear the shift into someone else's phrasing, and the first follow-up breaks it.
• Never doing a human mock. AI won't frown, interrupt or check its phone, and real interviewers sometimes do all of that.

How does AuraCloak help during practice sessions?

AuraCloak is our Windows app: an always-on-top AI chat overlay that sits over whatever you're working in. For mock interviews, a few features do the work.

Put the brief (prompt 1) in the custom system prompt. That matters more than it sounds: AuraCloak sends the last 10 exchanges as conversation context, so in a long session the original brief would scroll out of a normal chat, but the system prompt is always included. Answer out loud with push-to-talk: Alt+V starts recording your microphone and sends it when pressed again, and Alt+Shift+V cancels a false start. With the Local Whisper engine (a 141 MB or 465 MB model), your practice audio stays on your PC.

Alt+S screenshots your screen into the chat, which is handy for a system design diagram or a coding problem you're practicing. Choose the model, from OpenAI, Claude and Gemini to a local one through Ollama, or skip API keys and use browser mode with your existing ChatGPT Plus or Claude Pro login (there, paste the brief as your first message, since the system prompt applies to API providers). There's no built-in question bank or scoring, so the prompts above supply the structure. The interview prep use case shows a full session.

Comparing paid practice tools as well? See AuraCloak vs Final Round AI and AuraCloak vs Verve AI, or the wider list of AI interview assistants.

FAQ

Can AI do a realistic mock interview?

For question practice, yes. A voice model given the job description and your resume asks relevant questions and follows up when told to. It's weaker than a person at unexpected follow-ups and body language, and it praises too easily unless you ask for strict scoring, so add at least one human mock.

What is the best free AI mock interview tool?

Gemini Live and ChatGPT's voice mode both run spoken mock interviews on free accounts, with usage limits the vendors don't fix publicly. Yoodli's free plan includes 5 lifetime roleplay sessions with delivery feedback, and Verve AI's free plan includes 5 mock interviews.

Is Google Interview Warmup still available?

No longer at its old address. When we checked on October 14, 2026, grow.google/interview-warmup redirected to a general interview tips article. Google's Coursera course on interview practice with AI now uses Gemini Live for mock interviews instead.

How many mock interviews should I do before a real interview?

There's no fixed number, but three or more full spoken mocks for the round you're facing is a good floor, with your weakest answers re-drilled between them. Start one to two weeks before a big interview loop.

Does Final Round AI's mock interview cost money?

Yes. Final Round AI's free plan covers goals and a prep checklist, but practice interviews are listed under Pro, which started at $25 a month on its pricing page in October 2026. Pro includes eight practice scenarios and automatic debriefs.