Mock Interviews: AI Voice Practice, Peers and Recordings

Many students prepare for interviews by reading solutions and typing answers. Then, in the real interview, they find that they cannot explain an approach they understand, that their voice trails off when they are unsure, and that a single "why?" from the interviewer scatters their thinking. None of that showed up while practising, because practice was silent, written and private.

The fix is to rehearse the way interviews happen: aloud, live, timed, with someone who asks follow-up questions. Today there are three practical ways to get that, and they work best together.

• How to use AI voice modes for mock interviews, and their limits.
• How to set up useful peer and senior mocks.
• How to record, transcribe and analyse your own practice.
• A week-before plan that uses all three.

Several assistants offer voice modes on phones and computers (for example Gemini Live and ChatGPT's voice mode, as of October 2026, on both Android and iOS). You talk; it answers aloud; you can interrupt. That makes it a reasonable stand-in for an interviewer in discussions.

Start the session by speaking or typing an instruction like this:

You are interviewing me for an entry-level software engineer role for 40 minutes.Give me one medium problem on [arrays / strings / graphs]. I will explain myapproach aloud and describe my code as I write it in my own editor.Rules: do not give hints unless I have been stuck for several minutes or ask.If I go quiet for long, remind me to think aloud. When I state a complexity, ask meto justify it. At the end, give me feedback on clarifying questions, approach,communication, testing and complexity analysis, one line each.

Write the code in your own editor, as you would in an interview, and narrate it. Afterwards, run the code yourself on test cases; do not rely on the assistant's verdict about whether it works.

The most valuable use of AI voice is making you defend your answers, which most practice never does:

Now act as a sceptical senior engineer. Challenge my solution: ask me to prove thecomplexity, ask what happens on [edge case], ask why I did not use [alternative],and ask what breaks if the input were 100 times larger. Be precise and professional,not rude. Keep pushing until I either defend an answer clearly or admit a gap.

Saying "I'm not sure; let me reason it through" out loud, calmly, is a skill. This builds it.

Voice works especially well for behavioral rounds (Module 9) and design discussions (article 32), which are conversations anyway. Use the follow-up drill from article 36 and the design prompt from article 32 in voice mode.

• Assistants can be wrong about code, complexity and facts, and sometimes agree too easily. Verify technical points yourself.
• An AI does not reproduce a real interviewer's unpredictability or reactions.
• Voice conversations may be stored by the provider. Do not share personal details or a company's confidential interview content.

A real person notices things an AI does not: that you avoid eye contact when unsure, that your explanation was clear but too long, that you seemed defensive about a hint.

Pair with one or two friends who are also searching:

• Swap roles. Each session, one interviews and one is the candidate, then switch. Playing the interviewer teaches you what good answers look like from the other side.
• Use a real format. 45 minutes, a problem the candidate has not seen, a shared editor or document, the interviewer staying mostly quiet.
• Give structured feedback on five areas: clarification, approach, communication, testing, complexity. One strength and one improvement each.
• Be honest. Kind but specific feedback is the whole point.

People who recently cleared interviews at your target companies (article 14) are the best mock interviewers you can get. Ask for one 45-minute mock, not ongoing coaching, a week or two before an important interview, and make it easy: send the format and a time, and offer to share the problem or let them choose.

Some platforms match you with peers for free mock interviews, or with experienced engineers (often paid). Pramp, a well-known free peer-matching service, became part of Exponent, whose peer practice continues under the Exponent name; interviewing.io offers mock interviews with engineers from larger companies. Offerings and prices change, so check the current terms.

Recording is the fastest way to see yourself as an interviewer does.

1. Record a practice session (your phone's voice recorder, or the recording feature of a video call with only yourself). If you record a session with another person, ask their permission first.
2. Transcribe it (many phones and note apps can transcribe; transcription tools have free tiers).
3. Analyse the transcript with an assistant:

This is a transcript of me solving a coding problem aloud in a mock interview.Analyse it:1. Count filler words (um, uh, basically, like, actually, you know, sort of).2. Did I restate the problem and ask clarifying questions before solving?3. Did I state my approach and its complexity before coding?4. Find gaps where I said nothing useful for a long stretch (look for jumps in   the timestamps).5. Did I test with an example and an edge case before finishing?6. What is the single most important change for my next session?Quote the transcript for each point.[paste transcript]

4. Fix one thing at a time. Next session, focus only on that change. Re-record and compare.

Speech-coaching apps that score pace and filler words also exist; they can be useful for behavioral answers, but the prompt above covers most of what a fresher needs.

Not every interview is a coding round. Adapt the mock to the format you expect.

For any of these, the transcript analysis prompt above works with small changes: ask about clarity, structure and whether each answer connected to a concrete example.

While interviews are coming:

• Only silent practice, then a first live explanation in the real interview.
• Trusting AI's verdict on your code without running it.
• Peer mocks with no structure or honesty.
• Never listening to your own recordings.
• Trying to fix everything at once after a mock.
• Exhausting yourself with mocks the day before.

You are interviewing me for 30 minutes for an entry-level role. Give me onemedium problem that needs binary search, possibly binary search on the answer.Rules:1. Do not reveal the pattern; let me find it.2. No hints unless I ask or have been stuck for several minutes; then give the   smallest useful hint.3. When I describe my loop, ask me what the loop invariant is and how I avoid an   infinite loop.4. Ask me about edge cases: empty input, all equal values, the answer at either end.5. At the end, one line of feedback each on clarification, approach,   communication, testing and complexity.

Session 1 (date): filler words 31; stated complexity only at the end; one 70-secondsilence while debugging; traced example but no edge case.One change: state approach and complexity BEFORE coding.Session 2 (date): filler words 24; approach and complexity stated at minute 6;silences under 30 seconds; still no edge case.Next change: test one edge case before saying done.

• Next: Your Experience Inventory and Story Bank, Without an Internship prepares the behavioral round.
• For practising interviews with AI in more depth, see Interview Preparation and Mock Interviews with AI.
• For a full preparation timeline, see Software Interview Preparation: A 6-Week Plan That Actually Works.