Yes, you can use ChatGPT for a mock interview. Give it your target role, a short job description and verified experience notes, then ask it to interview you one question at a time. Answer before requesting coaching. At the end, ask for feedback tied to your actual words and repeat the weakest answer without reading a script.
Below is a copyable prompt and an original worked example showing the whole practice loop: a vague answer, follow-up questions, a scoring rubric and a stronger retry. You can start with text; voice is optional.
Method: This is a research-based tutorial checked November 12, 2026. The sample candidate, dialogue, numbers and scores are fictional editorial teaching material, not a captured ChatGPT session or a product benchmark. ChatGPT may respond differently. SkillCopilotAI publishes this guide and sells interview assistance; its relevant trial is discussed after the exercise.
This independent guide is published by SkillCopilotAI and is not affiliated with, sponsored by, or endorsed by OpenAI.
Cover: AI-generated editorial illustration; it does not show a real interview or product session.
Set up one useful practice session
Pick a role and an interview stage. “Entry-level financial analyst, hiring-manager behavioral round” gives the exercise a clearer target than “interview me.” For a technical round, specify the language, topic and level rather than expecting the model to know your employer's private questions.
Prepare a short context packet:
- Role and requirements: Paste the relevant responsibilities from the job posting, distinguishing requirements from preferences.
- Your evidence: Include two or three real projects, what you personally did, one decision and its outcome. An internship, coursework or volunteer project can work if you describe it honestly.
- Practice target: Choose a problem such as explaining your contribution, defending a decision or answering follow-ups without rambling.
Remove phone numbers, addresses, client names and confidential work details that the exercise does not need. Use placeholders or a fictional practice scenario when you cannot share the underlying facts. OpenAI's job-application guidance similarly recommends verified evidence and checking interview examples against your experience.
Open a new ChatGPT conversation and paste the packet with the prompt below. Pasting a short text summary avoids making file-upload access a prerequisite. Before beginning, ask it to summarize the role and the facts it may use; correct any misunderstanding.
Copy this ChatGPT mock interview prompt
Replace the bracketed fields with your own information. These instructions are requests to the model, so check that it follows them.
Run a practice interview for [role, seniority, industry and interview stage].
Job requirements: [paste relevant job-description text].
Verified experience notes: [paste your own facts, actions and outcomes].
Practice target: [one skill to improve].
Interview rules:
1. Ask five main questions, one at a time, and wait for my answer.
2. After each answer, ask up to two follow-ups about a missing fact,
my personal contribution, or the reason for a decision. Wait each time.
3. Do not supply a model answer or coach me during the interview.
4. Ask about the supplied role requirements. Do not claim to know the
employer's actual interview questions or private hiring rubric.
5. Do not invent achievements, metrics or experience for me.
6. Stop early if I type DEBRIEF. Otherwise debrief after question five
and its follow-ups. I will manage the session time myself.
During the debrief, assess answer content using four criteria:
relevance, personal ownership, reasoning/trade-offs, and outcome evidence.
For each criterion use 0 = absent, 1 = present but vague or incomplete,
2 = specific and supported by what I actually said.
Quote the evidence for each score. Mark unsupported claims or uncertain
facts; do not fill gaps. Do not score my voice or body language from text.
List one priority to improve, then ask me to retry the weakest answer.
Do not predict whether I will be hired.
Start by briefly confirming my role/context, then ask only question one.Check the first response: it should confirm the context and ask one question. If it produces all five questions or answers them for you, use the repair instructions below. For a short first attempt, answer one main question and its follow-ups, then type DEBRIEF.
Use a separate timer if you want a 15- or 20-minute session. The prompt's question count is a stopping rule you can inspect; it does not make ChatGPT a reliable stopwatch.
Prefer speaking? Use voice or dictation deliberately
Text is convenient for inspecting an answer's structure. To rehearse speaking, say the answer aloud before sending it, or use voice when it is available in your ChatGPT account.
For the desktop experience documented in OpenAI's ChatGPT Voice guide, start a new chat, select the voice control and allow microphone access when prompted. State the role, context and one-question-at-a-time rules before beginning. That guide lists Plus, Pro, Business, Edu and Enterprise availability, with rollout and workspace conditions. Check your own app and plan rather than assuming every account or device has identical access or limits.
Voice is a live conversation. Dictation turns speech into prompt text before you send it. With dictation, inspect the transcription for names, numbers and technical terms; correct recognition errors without rewriting your answer into polished prose. Otherwise the feedback will assess a different answer from the one you spoke.
Use a recording you make and review yourself, or a human practice partner, to check delivery. A written transcript alone cannot establish your eye contact, pace, tone or confidence.
Worked example: a data analyst explains a process improvement
This is an original fictional mini-round, including the interviewer lines and editorial scoring. It demonstrates how to inspect feedback; it is not a transcript generated by ChatGPT. It covers one question and two probes, rather than claiming to reproduce a full five-question interview.
The candidate's fact sheet
The example role requires SQL, data validation and explaining results to nontechnical colleagues. The candidate has these practice facts available:
- A weekly revenue report reached the Monday meeting late because finance had to reconcile duplicate totals.
- Joining orders to multiple payment records duplicated some orders. The candidate aggregated payment records by order before joining.
- The candidate compared a 20-order sample against source records and reviewed exceptions with a finance analyst.
- The corrected report was available by Monday's 9 a.m. meeting for three consecutive weeks. There is no verified revenue-growth or hours-saved figure.
These numbers belong only to the fictional exercise. Replace them with facts you can defend; do not borrow them for your resume.
Question, initial answer and two follow-ups
Interviewer: “Tell me about a time you improved a reporting process. What did you change, and how did you know it helped?”
Candidate: “Our weekly revenue report was late. I worked with the team to update the SQL and checked the numbers. After that, the report usually arrived earlier and finance was happier with it.”
The answer is relevant, but “worked with the team” hides ownership and “finance was happier” provides little evidence.
Interviewer: “Which part did you personally change, and why was it causing the problem?”
Candidate: “I found that joining orders to individual payment rows repeated orders with multiple payments. I changed the query to aggregate payments by order before the join. A finance analyst reviewed the exceptions with me; I didn't make the accounting decisions myself.”
Interviewer: “How did you check the fix, and what outcome can you support without estimating?”
Candidate: “I checked 20 orders against the source records, including orders with multiple payments, and reviewed mismatches with finance. The report was ready for the Monday 9 a.m. meeting for the next three weeks. I don't have a measured time-saving or revenue figure.”
The probes expose a query decision, a validation method, a collaborator's role and a bounded result. None requires inventing a percentage improvement.
Score the answer against evidence, then retry
Use the same four criteria from the prompt. A 0 means the element is absent, 1 means it is present but incomplete, and 2 means it is specific and supported within the answer. This is our teaching rubric for answer content, not an employer's rubric or a validated measure of interview readiness.
The following assessment scores the initial answer alone, before the probes. It shows what was missing from that answer, even though the candidate knew more.
| Criterion | Initial score | Evidence and missing detail |
|---|---|---|
| Relevance | 2/2 | Identifies the late revenue report and a process change. |
| Personal ownership | 1/2 | “I worked with the team” does not identify the candidate's change. |
| Reasoning and trade-offs | 0/2 | Gives no cause, reason for the SQL change or validation decision. |
| Outcome evidence | 1/2 | “Usually arrived earlier” describes an outcome without a checkable observation. |
| Total | 4/8 | A revision target for this exercise, not a hiring prediction. |
Priority: Bring the cause, personal action and verified result into the first answer. Do not make the listener extract all of them through follow-ups.
Ask for a retry before requesting a model answer. Here is a revised version assembled only from the fictional fact sheet and follow-up answers:
Our weekly revenue report was reaching the Monday meeting late because finance had to reconcile duplicate totals. I traced the duplication to a join between orders and individual payment records. I aggregated payments by order before joining, then checked 20 orders against source records, including multiple-payment cases. I reviewed exceptions with a finance analyst, who handled the accounting decisions. The report was ready by the Monday 9 a.m. meeting for the next three weeks. I can support that delivery result, but I haven't measured hours saved or revenue impact.
Under this narrow rubric, the revised answer earns 2/2 on each criterion, or 8/8: it stays on the reporting task, identifies the candidate's SQL change, explains the duplicate-row cause and validation, and states an observable result. That does not prove the SQL was correct, the sample was representative or the candidate is ready for the job. A technical interviewer could still ask how refunds, partial payments or missing records were handled.
For your own retry, hide the written version and answer aloud in your own words. Then try a new question: “Tell me about a time a stakeholder challenged your analysis.” If you can only deliver the rehearsed paragraph, you have practiced recall more than adapting an explanation.
Fix common mock-interview problems
| What happens | What to do next |
|---|---|
| It gives a list of questions | Send: “Ask only the first question and wait. Keep the remaining questions for later.” |
| It writes your answer for you | Send: “Stay in interviewer mode. Ask me for the missing detail before offering wording.” |
| Feedback says “be more specific” | Ask it to quote the vague phrase and name the fact needed. Supply the fact yourself. |
| A score seems generous or changes | Ask for evidence under each unchanged criterion. Challenge unsupported points; keep your own rubric and answer log. |
| It adds a result you never supplied | Remove the invented detail. Say: “Use only my fact sheet; mark the outcome unknown if it is missing.” |
| Voice recognition changes your claim | Correct the transcript first and request a fresh assessment of the corrected answer. |
| It loses context or repeats questions | Start a fresh chat with the same short packet, rules and the last completed question. |
These recovery steps respond to problems users have reported, including turn-taking and feedback concerns in the original mock-interview discussion. Those older reports are useful warnings, not evidence that every current session will fail in the same way.
Save the question, unedited first answer, follow-ups, criterion evidence and unaided retry in a private practice log. When a score rises, identify the change in the answer that justifies it. For a longer routine, use our seven-day AI mock interview practice guide.
When SkillCopilotAI is worth trying
Stay with ChatGPT if this setup gives you useful questions and concrete feedback. You do not need another subscription to follow the exercise if your existing ChatGPT access supports the text conversation.
If you want to evaluate a dedicated desktop interview workflow, SkillCopilotAI offers AI mock interview practice and resume/target-role context for answer guidance. Our pricing page lists a 60-minute trial with limited model access, plus unlimited mock interview practice during a paid plan's active term. That documented feature list does not establish better feedback than ChatGPT, automatic consistent scoring or a particular interviewer's behavior.
In a private trial, use the same role and one real project from this exercise. Check whether the questions or guidance use your facts accurately, whether you can work through a useful follow-up, and whether you can explain the answer afterward with assistance closed. If it supplies unsupported achievements or does not help your actual practice gap, keep using the simpler workflow.
Use a supported Windows or macOS computer; check the downloads page for the current system requirements. After installing and signing in, start with the mock-interview practice option rather than treating live answer suggestions as your own unaided performance. Our mock interview vs interview copilot guide explains that distinction.
Questions before you start
Can I do a ChatGPT mock interview for free?
Use the text workflow if it is available within your account's current allowance. This guide does not require a specific paid model or file upload. Voice, usage limits and other features depend on your account, app and current availability; check those before planning a long session.
Can it ask technical or behavioral interview questions?
You can request either. For behavioral practice, provide true project or experience notes. For technical practice, name the topic and level, and check factual feedback against authoritative documentation or by running the relevant code. Treat generated questions as practice material, not a leaked employer question bank.
Should feedback come after every answer or at the end?
Use feedback after each answer when learning a specific skill. For a rehearsal, wait until the end so coaching does not shape your next response. The prompt above uses end-of-session feedback; the worked example debriefs after one mini-round.
How do I know the practice helped?
Answer a new, related question without reading generated wording. Check whether you can explain your contribution, reasoning and real outcome, then handle a follow-up without adding facts. Ask a human partner to review a recorded attempt when delivery or technical judgment remains uncertain. A chatbot's score alone cannot answer that question.
Choose your next step
Compare tools or explore guidance for a permitted practice conversation.
