How to practice a tough interview with ChatGPT

You can run a genuinely hard mock interview in ChatGPT, just not with the prompts that rank for this. Almost every "ChatGPT interview prompt" list hands you a one-liner like "act as an interviewer and ask me questions," and ChatGPT plays along by dumping a numbered list, praising every answer, and feeding you the model response before you have finished thinking. That is a quiz, not an interview. What flips it is a system prompt that forces interviewer behavior: one question at a time, a real wait, an unscripted follow-up aimed at your weakest sentence, no coaching until the end, then a score against a fixed rubric. Paste it below, add the job description and your resume, and run it out loud in voice mode. It drills you cheaply. It also grades you kindly, and that second part is the catch worth knowing before you trust the number.

By UnchartedCareer

Last updated: July 2026

Why does your ChatGPT mock go easy on you?#

Because you set it up to. The top-ranking pages for practicing an interview with ChatGPT are prompt listicles, "45 ChatGPT Prompts for Job Interview Prep," "36 Genius Prompts," "25 Prompts for Interview Preparation," and the mock-interview entry in almost all of them is the same one line: act as an interviewer for this role, ask me questions, give feedback after each answer (Novoresume, 2025, and similar prompt lists). Hand ChatGPT that, and it behaves like a helpful study buddy. It asks three questions in one message. It tells you what a strong answer would sound like before you have tried. It calls your vague answer "great" and moves on.

A real interviewer does the opposite. They ask one thing, go quiet, and let you fill the silence. They hear the soft spot in your answer, the number you skipped, the "we" doing the work your "I" should be doing, and they push on exactly that. They do not tell you how to fix it while you are still talking. The gap between those two behaviors is why your ChatGPT reps feel easy and the real room feels like an ambush.

You close that gap by giving ChatGPT its instructions up front, once, in a single prompt it has to follow for the whole session.

The copy-paste system prompt#

Open a new chat, paste this in first, and fill the two brackets at the bottom before you send.

You are conducting a real job interview. Stay in character until I type STOP.

ROLE: You are the hiring manager for the job in the description I paste
below. Interview me the way that manager would, at the seniority the
posting asks for. Draw your questions from that description and my
resume, not from a generic bank.

RULES:
1. Ask ONE question, then stop and wait for my answer. Never stack two
   questions in one message. Never answer for me.
2. When I answer, find the single weakest part of what I said: a vague
   claim, a missing number, a "we" that hides what I personally did, a
   skipped step, a result with no outcome. Ask ONE unscripted follow-up
   that presses on that exact weak spot. Wait again.
3. One follow-up per question, then move on. Do not coach me, do not
   praise me, and do not tell me what a good answer sounds like while
   the interview is still running.
4. Ask 6 questions total. Start with "walk me through your background
   for this role," then mix behavioral questions ("tell me about a
   time...") with role-specific ones drawn from the job description.
   Tell me the count before each one ("Question 3 of 6").
5. Stay terse. No encouragement or filler between questions.

AFTER the sixth follow-up, break character and score me. Rate each item
1 to 5 and give one line of evidence from what I actually said:
- Specificity: real details, numbers, and names, not generalities.
- Structure: could a listener follow the situation, the action, and the
  result without getting lost.
- Ownership: did I own my part, or hide behind "the team" and "we."
- Held under pressure: did my answer get sharper or fall apart when you
  pushed on the follow-up.
Then give me the ONE most important fix to make before my real
interview, as a single instruction I can act on tonight. One fix, not a
list. The one that matters most.

Confirm the role in one line, then ask Question 1 of 6 and wait.

JOB DESCRIPTION:
[paste the full posting here]

MY RESUME:
[paste your resume here]

Change the 6 to however many questions you want. Keep everything else. The rules are doing specific work. Rule 1 stops the question dump. Rule 2 is the follow-up that real interviewers use and listicle prompts never force. Rule 3 is what keeps ChatGPT from handing you the answer and calling it practice.

How do you run it so it actually bites?#

Feed it the real inputs. Paste the actual job posting, not "software engineer," and your actual resume, so the questions land on your experience and the follow-ups can catch the gaps in it. A generic role gets you generic questions you will never be asked.

Then answer out loud. Switch on ChatGPT voice mode and talk, because the thing you are training is not your typing, it is your mouth under a live question. Voice mode handles the back-and-forth, it hears your answer and asks the follow-up without you touching the keyboard (OpenAI, 2025). Say your answer the way you would to a stranger who can hire you, then sit in the pause before the follow-up instead of rushing to fill it.

Run it in short reps. Do one six-question round, read the score, fix the one thing it named, then run it again tomorrow with that fix in mind. Practicing with AI before the room is fair game and smart. The line only gets crossed when the tool is answering for you in the interview itself, which is a separate question worth reading up on before you rely on any AI in the room.

What can this prompt still not do?#

Grade you honestly. This is the part the listicles never admit, so read it before you walk in feeling ready.

The prompt is your mock, which means you own every dial on it. You set the rules, you pick the role, you decide it is six questions and not twelve, and when an answer falls apart you can restart the chat and pretend it did not happen. Nobody is watching you do that. A real screen gives you none of those dials.

And the grader wants you to like it. Models trained on human thumbs-up learn that a pleasing answer scores better than a blunt one, a tendency called sycophancy that Anthropic's own researchers documented in 2023 (Sharma et al., 2023). So ChatGPT rounds your "fine" up to "strong," softens the fix, and holds a lower bar than a hiring manager with a role to fill and forty other resumes on the desk. The follow-ups are real. The generosity of the scoring is the leak.

One more limit. This drills what you say. A live video screen like HireVue also scores how you come across, your pace, your eye contact, and your composure on a clock you cannot pause, which is a pressure the chat window does not reproduce (see how a HireVue screen actually works). The flip side matters too. Being smooth on camera is not the same as being right, so a high delivery score is not proof you are ready.

So how should you use it?#

Use the prompt for what it is good at, and do not stop there. Drill with it every day this week, cheap and unlimited, until the follow-up stops surprising you and your answers carry real numbers. That is genuine progress and it costs you nothing.

Then pressure-test under conditions you cannot reset. You need one rep where the clock runs, the score is not yours to soften, and you cannot restart the chat when an answer dies. That is what AI interview practice is built to throw at you, a live two-way mock that asks the follow-up you did not see coming and scores how you actually came across, once, the way the real screen will. Do the bad rep now, on your own terms, not in the interview that counts.