Used well, a chat model is the rehearsal partner most people never had: available at eleven at night, happy to ask the same awkward question four times, and immune to being bored by your third attempt at the same answer.
Used badly, it writes you a polished paragraph about someone else's career that you then try to memorise. The prompts below are built to keep you on the first side of that line.
Give it the five things it needs
Almost every disappointing answer comes from a thin prompt. Before you ask anything, paste in the posting, your CV, the exact role title, a couple of lines about the company, and which stage you are at — recruiter screen, hiring manager, panel, final. Preparation for a first screen looks nothing like preparation for a final round, and without the stage it will default to generic.
Add one instruction that changes everything downstream: 'use only what is in my CV; if you need a detail I have not given you, ask me for it.'
Start by predicting the questions
Ask it to read the posting as an interviewer would and produce fifteen likely questions, grouped by type — behavioural, role-specific, motivation, and the awkward ones about your history. Then ask which three competencies the posting keeps circling back to, because those are what the panel has been briefed to score.
Follow it with a mapping prompt: which parts of my CV are most likely to be probed, and where would a sceptical interviewer push? That list is your prep plan, and it is usually shorter and more specific than the one you would have written yourself.
Turn rough notes into structured answers
The best use of a model here is not writing your stories but shaping them. Dictate the messy version — what happened, what you did, roughly how it ended — and ask it to restructure into situation, task, action and result without adding a single fact you did not supply.
Then ask the harder question: 'where is this answer weak?' A good response will tell you the setup runs long, the actions are described as a team's rather than yours, or the result has no number in it. Those three notes cover most weak answers anyone gives.
- 'Restructure this into STAR using only the facts I gave you. Mark anything you think is missing.'
- 'Cut this to ninety seconds without losing the measurable outcome.'
- 'Rewrite this so my own contribution is unambiguous — I keep saying we.'
- 'What follow-up questions would a sceptical interviewer ask about this story?'
Run the mock, then make it harder
Ask it to act as the hiring manager for this specific role and stage: one question at a time, waiting for your answer, asking a follow-up before moving on. The one-question-at-a-time instruction matters — without it you get a list of twenty questions, which is a document, not a rehearsal.
After a few rounds, raise the difficulty. Ask it to press on vague claims, ask for numbers you did not volunteer, and challenge one answer directly. That is where mock interviews start being useful, because the discomfort you feel is the discomfort you were going to feel in the room.
Refine, but keep your own voice
A useful refinement prompt asks for specifics, not polish: remove repetition, flag unsupported claims, cut hedging, and keep the phrasing in my own words. If what comes back does not sound like you, it is worse than what you had, however well written it is.
Say the final version out loud. Anything you stumble over is a sentence you would not naturally say, and it will sound recited under pressure.
Technical and role-specific rounds
For technical interviews, use it for breadth rather than depth: likely topics for this role and level, the trade-offs an interviewer will want you to articulate, and the fundamentals worth a refresher. Then practise the actual problems somewhere the answers are verified, because a model will state something wrong with complete confidence.
One prompt worth running whatever the role: 'given this posting, what would I be expected to know that a candidate from my background often does not?'
What it genuinely cannot do
It does not know this interviewer, this team's politics, or what went wrong with the last person in the seat. It will produce answers that are competent and slightly generic unless you push back. It cannot reproduce the pressure of a real room, which is a large part of why interviews are hard.
And it does not belong in the interview itself. Reading from a second screen is visible, it is recorded in a video round, and an interviewer who suspects it will ask exactly the follow-up you cannot answer.
Where the posting comes from
All of this starts with the right posting in front of you. Apply Engine scores roles against your profile before you spend an evening on them and shows which requirements you already meet — which is the same list your interview prep should be built around. The questions you get asked are usually the gaps it flagged before you applied.
