Practising interviews with ChatGPT works because it gives you the one thing that is otherwise hard to get: unlimited rehearsal with someone who has read the job description. Set up well, it asks realistic questions, follows up on vague answers, and scores you honestly. This guide shows how, step by step.
Everything here also works with Claude, Gemini or Copilot. For a bank of individual prompts, see our AI prompts for interview preparation; this article is about running the practice itself.
What AI mock interviews are good for — and not
| Good for | Not good for |
|---|---|
| Hearing likely questions before the real thing | Judging body language or eye contact |
| Rehearsing answers out loud until they are smooth | Reproducing the pressure of a real panel |
| Being pushed on vague answers | Knowing what this employer specifically values |
| Scoring structure, length and specificity | Replacing practice with a real person entirely |
Step 1: Prepare the material (10 minutes)
Gather three things as plain text: the job description, your resume — the exact version you sent — and a few notes on the company from its own website. Remove your address and phone number from the resume first.
Step 2: Set up the interviewer
Start a new chat and paste this, with your material underneath:
You are interviewing me for the role below. Act like a real hiring manager: ask one question at a time and wait for my answer. Mix questions about the role, my experience and behavioural "tell me about a time" questions. If my answer is vague, ask a natural follow-up, as a real interviewer would. Do not give feedback or praise during the interview. When I type "end interview", give me a score out of 5 for each answer and the three answers that most need work. Do not invent anything about my background; ask me instead. [job description] [resume] [company notes]
Two parts of that prompt do the heavy lifting. "No feedback during the interview" stops the assistant from congratulating you after every answer, which is its default and useless for practice. "Ask a natural follow-up" makes it behave like an interviewer rather than a quiz.
Step 3: Practise out loud — ideally by voice
Typing answers is better than nothing, but an interview is spoken. Typed answers come out shorter, tidier and more considered than anything you will say under pressure. Two ways to fix that:
- Use voice mode in the ChatGPT app, if it is available to you. You speak, it replies aloud, and you practise the real skill: answering in full sentences without a delete key.
- Or speak your answer, then type it — or use your device's dictation — so what you submit is what you actually said, fillers and all.
Either way, sit up, look at the screen, and answer as if it were real. The rehearsal is only as useful as it is realistic.
Step 4: Get honest feedback
After "end interview", assistants tend towards encouragement. Push for specifics:
Score each of my answers from 1 to 5 on: answered the actual question; specific rather than general; clear about what I personally did; included a result; right length for a spoken answer. Be strict — a 5 should be rare. Then show me my weakest answer rewritten using only facts I gave you.
That last clause matters. If the rewritten answer contains details you never said, the assistant is coaching you to claim things you did not do.
Step 5: Drill the weak spots
Most of the improvement comes from repetition, not from more full interviews. Take the two lowest-scoring questions and drill them:
Ask me "[question]" again. After each attempt, give me two things that improved and one thing to change. Keep going until I say stop.
Three or four attempts at one hard question will do more than answering twenty new ones. Behavioural questions usually need the most work; see how to prepare for behavioural interviews with AI and our STAR method guide.
Step 6: Turn up the pressure
Once your answers are solid, practise the uncomfortable version:
Run the interview again as a sceptical, time-pressed interviewer. Interrupt if I ramble, challenge claims that sound vague, and ask "what exactly did you do?" whenever I say "we". Stay professional.
It is better to meet this interviewer in a chat window than for the first time in a real room.
A three-session plan
| When | Session | Time |
|---|---|---|
| Three days before | Full mock interview (steps 2–4), note the weakest answers | 45 min |
| Two days before | Drill the weakest three questions (step 5) | 30 min |
| The day before | Pressure mock (step 6), then prepare your own questions to ask | 40 min |
On the day itself, do not practise. Reread your stories and your questions, and leave it there.
The line not to cross
Using ChatGPT to prepare is preparation. Using it during a live interview — reading answers from a screen, or having it listen and suggest replies — is misrepresentation, and it tends to show: delayed, oddly polished answers are exactly what interviewers learn to notice. Practise until you do not need it.
Practice modes, a scoring rubric and a practice schedule are the subject of a full chapter in Use AI to Get Hired, followed by chapters on STAR stories, common questions, technical interviews, and video and AI-assisted interviews, including where the ethical line sits. The free toolkit on the Use AI to Get Hired page includes a full mock-interview prompt and a video-interview setup checklist.
Frequently asked questions
Can ChatGPT do a mock interview for any job?
For most roles, yes, if you give it the job description. For specialist technical roles, add the specific skills you expect to be tested on so the questions are realistic.
How many mock interviews should I do?
One or two full mocks per interview, plus drilling. More full mocks bring diminishing returns.
Is AI interview practice as good as a real person?
It is more available and more patient, but a real person catches delivery issues AI cannot. If you can, do one final mock with a friend.