How to Use ChatGPT for Interviews (And Better Alternatives)
ChatGPT is genuinely good at interview preparation and genuinely bad at live interviews, and the reasons are structural rather than a matter of model quality.
This covers both halves: prompts that produce something useful beforehand, and what changes the moment the interview starts.
Where ChatGPT is excellent: preparation
Most people use it as a search engine — "common interview questions for X" — get generic output, and conclude it isn't useful. The value is in using it as a sparring partner instead.
Pressure-test your stories. Paste a STAR answer you've written and ask it to interview you about it:
Here's a behavioural answer I've prepared. Act as a sceptical interviewer and ask me four increasingly specific follow-up questions about it — the kind designed to find out whether I actually did this or I'm describing what my team did.
The follow-ups are where behavioural answers fall apart. Finding that out at your desk is much cheaper than finding it out in the room.
Generate questions from the actual job description. Paste the posting:
Based on this job description, list the twelve questions I'm most likely to be asked, ordered by probability, and say what each one is really testing.
Run mock system design. It's a decent design partner because it will push back:
Act as a senior interviewer. Give me a system design problem for a mid-level backend role, then challenge my design at each step. Don't give me the answer — make me justify choices, and change the requirements partway through.
Turn a project into quantified bullets. Feed it a rough description, ask for résumé lines with metrics, then supply the real numbers yourself.
Explain a concept at the level you'll be asked. "Explain consistent hashing the way I'd need to describe it in a 45-minute system design round, not the textbook version."
Rehearse the questions you ask them. Most candidates improvise this and it shows.
Why it stops working live
Four specific limitations, none of which are about how good the model is:
It doesn't hear the interview. ChatGPT has no access to your system audio. It can't hear the interviewer through your speakers. You'd have to type or paraphrase every question during the interview, which is exactly the thing you can't do.
It doesn't see your screen. In a coding round the problem is text in a browser panel. ChatGPT can't read it unless you copy and paste it in — visibly, while someone is watching you.
It's a visible window. A browser tab or desktop app that appears in your dock, your window switcher and any full-screen share. It's designed to be looked at.
Everything requires an action. Typing, pasting, clicking, reading. Each is a visible behaviour on camera, and on assessment platforms typing and pasting may be logged.
The result is a tool that answers well and can't be reached at the moment you'd need it.
Feature Spotlight: What changes with a purpose-built tool
The category exists to close exactly these four gaps. Velin AI transcribes system audio so it hears the interviewer directly, reads what's on screen with capture mode so nothing is pasted, runs hidden at the OS level rather than in a visible window, and can be driven by mouse gestures so there's no keystroke to log. Same underlying model quality — the difference is entirely reach and visibility.
If you're using ChatGPT live anyway
Some people will, so the practical version:
- Set it up beforehand with your résumé, the job description and your story grid already in the conversation, so you're not typing context under pressure.
- Share a window, not your entire screen. Window shares exclude everything behind them; full-screen shares don't.
- Turn on Do Not Disturb, especially if you're logged into it on your phone.
- Don't paste code from it into an assessment platform. Paste events are commonly logged, and a solution arriving in one block looks nothing like typing.
- Expect the follow-up to catch you on anything you didn't understand before reading it out.
The honest comparison
| | ChatGPT | Purpose-built assistant | |---|---|---| | Hears the interviewer | No | Yes — system audio transcription | | Reads your screen live | No | Yes — capture mode | | Visible on screen share | Yes | Designed not to be | | Requires typing or pasting | Yes | No, if it follows the conversation | | Roughly | ~$20/mo | $34/mo, or $324 once for Velin AI | | Best for | Preparation | The interview itself |
A Word of Caution
Relying solely on a tool to provide answers you don't understand is a risky strategy. True professional growth comes from learning, not just passing a test.
A practical point alongside that: an answer you don't understand is a liability regardless of where it came from. Interviewers ask follow-ups specifically to find the boundary of what you know, and a fluent answer followed by a blank on "why did you choose that?" is worse than a slower answer you can defend.
What to actually do
Before the interview, ChatGPT is one of the best tools available and most people underuse it. Pressure-test your stories, generate questions from the real job description, rehearse system design against something that argues back.
During the interview, it isn't built for the job — it can't hear the room, can't see your screen, and can't be reached without visible effort. Either go in on your preparation, or use something designed for the moment.
Related: the complete guide to AI interview assistants on how that category works, the roundup for the options ranked, and what your screen share actually shows.