Voice AI Mock Interviews: How to Rehearse With ChatGPT, Gemini and Claude

Job candidates now have an unusual coach available around the clock: the voice mode of a frontier AI assistant. ChatGPT, Gemini and Claude can all hold a spoken conversation, ask follow-up questions and tell you which answers sounded thin. The catch is that the same tools are increasingly off-limits once the real interview begins.

  • Three viable options: OpenAI lists interview preparation among ChatGPT Voice’s uses, Google’s careers guidance points to Gemini Live, and Claude’s voice mode is in beta on every plan, including the free tier.
  • Prompting decides quality: supply the job description, the interview format, turn-taking rules and the kind of critique you want, or you will get generic questions and empty praise.
  • Privacy matters: Gemini can retain audio recordings when History is on, while Claude converts speech to text and deletes the audio by default.
  • The real interview is different: Anthropic and Amazon both bar generative AI during live interviews without explicit permission.

Which assistant actually works for a spoken rehearsal

All three are serviceable, and their makers say so themselves. OpenAI cites interview preparation among the uses of ChatGPT Voice and suggests asking it to assess a presentation speech in a direct and constructive way. The full transcript can be reviewed by tapping the on-screen orb.

Google has moved further. The address of its old Interview Warmup tool now redirects to a Grow with Google interview-preparation guide dated 11 December 2025, which recommends Gemini Live for simulating a real-time conversational interview with structured feedback, and for practising the STAR method (situation, task, action, result). According to Gemini Live’s help documentation, the feature runs in the mobile app rather than the web version.

Claude’s voice mode is in beta across all plans, including the free one, on mobile, desktop and web, though Anthropic concedes it works best from a phone. Non-English languages are also in beta, and the spoken language is selected separately under Voice > Spoken language in the capabilities settings; changing the app language is not enough. Voice conversations count against your plan’s usage limits.

If pauses make you nervous, Claude’s push-to-talk mode stops the assistant from cutting in while you think. No real interviewer hands you a button, though, so the final rounds are best rehearsed hands-free.

How to prompt a rehearsal that is actually useful

The gap between a productive session and a flattering chat lies in the opening instructions. A chatbot with no context asks generic questions and applauds every answer. Before starting, give it four things:

  • The role and the posting, with duties and requirements copied from the advert, minus any non-public company data.
  • The format: HR screen, technical round or hiring-manager interview, and how many questions you want.
  • The rules: one question at a time, wait for you to finish, probe when an answer is vague, and withhold the model answer until the end.
  • The critique style: ask it to flag filler words, answers without numbers, evasions and moments where you failed to address the question.

A workable opener: “You are a recruiter hiring a junior data analyst. Ask me eight questions, mixing behavioural and technical, one at a time. If I answer in generalities, follow up asking for a concrete example. At the end, score each answer using the STAR method and tell me the three I most need to work on.”

One limit is worth keeping in mind: the model does not know what a given company values or how it interviews. Its feedback covers structure and clarity, not fit. And as with CV writing, do not ask it to invent achievements to fill out the STAR frame. Anthropic cites exactly that as a prohibited use in its candidate guidance: asking Claude to write application answers produces generic content built on experiences you never had.

What you should never tell the chatbot

Nothing you would not tell a stranger taking notes. In a rehearsal it is tempting to supply real context: your current salary, why you are leaving, your manager’s name, a confidential project. None of it improves the practice, and all of it is stored.

In Gemini, the app privacy hub explains that with the History setting enabled, audio recordings from Gemini Live are retained too, and activity is deleted by default after 18 months. With history off, conversations are kept for roughly two months. Human reviewers examine a portion, and what they review is retained for up to three years even if you delete it. Google’s summary is blunt: do not include confidential information you would not want a reviewer to see.

In Claude, the documentation states that audio is not retained by default: speech is converted to text, the recording is deleted, and the transcript stays in your history as a written chat.

The practical rule: discuss achievements with approximate figures and no proper nouns, and if you rehearse a salary negotiation, use public sector ranges rather than your actual payslip.

Can you use AI during the real interview?

It depends on the employer, and the default answer is no. Anthropic’s guidance, updated on 10 July 2025, separates the stages. It encourages using Claude to research the company, practise answers and prepare questions, but take-home assessments are to be completed without AI “unless we state otherwise”, and live interviews permit no AI assistance without authorisation. Candidates are also asked to be transparent about following these rules.

Amazon takes the same line. According to internal guidelines seen by Business Insider and reported in March 2025, the company asks candidates not to use generative AI tools during interviews without explicit permission, warns that violations can lead to disqualification, and lists signals for interviewers such as typing while answering or reading rather than speaking naturally.

The boundary these employers are drawing looks reasonable and likely to spread: AI as a coach beforehand, not a teleprompter during. A well-built rehearsal forces you to order your examples, time your answers and hear yourself say things that looked clear on paper. A prompt in your ear does the opposite, turning you into someone who reads aloud, which shows by the second follow-up. If you are unsure what a particular employer allows, ask the recruiter before the interview; it is a question that reflects better on you than the alternative.

Sources & References