How to use AI in a job interview: the full 2026 guide

Three years ago, bringing AI into a job search was a novelty. In 2026 it is standard practice on both sides of the table. Companies rank resumes with algorithms, run first screens with voice AI recruiters, and switch on screen recording or proctoring for technical rounds. Candidates answer in kind: they pull job posts apart with an LLM, rehearse out loud with voice models, and some bring a real-time assistant to the call itself.

"Using AI for interviews" is not one thing. It is three different jobs, and they carry very different levels of risk:

• Preparation. Breaking down the job post, predicting questions, sharpening your resume and your story. Useful to everyone, and the least contentious of the three.
• Practice out loud. Mock interviews with a voice model, working on phrasing, working on a second language. Also low risk, and the stage that moves the needle most.
• Live help. An AI assistant that listens during the interview and puts a prompt on your screen. It works in online formats, but it needs a clear head about the rules and the limits.

Here is how to run all three.

How to prepare for an interview with AI

At this stage an LLM does the job of a career coach, for free. Any strong model works: ChatGPT, Claude, Gemini, DeepSeek. The gap between models is much smaller than the gap between a good prompt and a lazy one.

Take the job post apart

Start by pasting the full job description in and asking the model to decompose it. A good model pulls out what your eye skims past: requirements that repeat in different words, the employer's real priorities, hints about the stack and how the team is set up.

A prompt that works: "Here is a job description. List the 10 most important requirements in descending order. For each one, tell me the question an interviewer would use to test it and what they want to hear in the answer." What you get back is a map of the interview before it happens.

Predict the questions, then answer them yourself

Next, ask for a list of likely questions for that exact role and level: behavioural ("tell me about a conflict on your team"), technical, case questions, motivation questions.

One rule matters here more than any other: do not ask for the answers. Write your own, in full sentences, and put the model in the critic's seat. Ask it to hunt for filler, missing specifics and missing numbers. That way you end up with better versions of your own answers instead of memorised generic ones.

Build your "tell me about yourself"

Almost every interview opens with it, and almost everyone improvises. Give the model your resume and the job post, and ask for a sixty to ninety second story built as present, past, future: what you do now, what led you here, why this role is the logical next step.

Then rewrite the result in your own words. A script that was memorised word for word sounds exactly like a script that was memorised word for word.

Tailor the resume to one job

Models are genuinely good at re-pointing a resume at a specific posting: reordering emphasis, translating your work into the employer's vocabulary, flagging weak spots like "this whole block has no numbers in it, add one measurable result."

The rule is simple: AI edits facts, it does not invent them. Borrowed experience falls apart on the first follow-up question, and that is a worse outcome than a modest resume. If you are changing fields, the same rule decides everything: frame what you actually have in the language of the role you want, instead of borrowing what you do not have.

Five prompts to start with

• Job post breakdown: "List the 10 key requirements in this job description in descending order of importance, plus the question that would test each one."
• Question bank: "Write 15 likely interview questions for a [role and level] in [industry]: behavioural, technical, case and motivation."
• Answer critique: "Here is my answer to question X. Find the filler, the vagueness and the places missing numbers. Suggest how to make it stronger."
• Your story: "Here is my resume and the job post. Build a 90 second 'tell me about yourself' as present, past, future."
• Weak spots: "Compare my resume with this job post. Name the 5 requirements where I look weakest and how to handle questions about them."

AI mock interviews: the part people skip

There is a canyon between knowing an answer and saying it out loud under pressure. Written prep does not cross it. Speaking does, and voice modes changed the economics of that completely: a full rehearsal is now available at any hour, with no partner needed.

Run it as a role play

Give the model a character: "You are a tech lead interviewing me for a senior QA role. Ask one question at a time. After each answer, dig deeper the way a real interviewer would. At the end, tell me what landed and what was weak."

Answer out loud, as if you were on the call. Three or four run-throughs take a real bite out of interview nerves, because your brain stops reading the format as a threat.

Get scored on structure, not vibes

Ask for judgement against explicit criteria: STAR structure for behavioural questions (situation, task, action, result), a ceiling of about ninety seconds per answer, filler words. A model is unembarrassed. It will tell you flatly that you spent four minutes answering a different question than the one you were asked. A friend doing you a favour usually will not.

Track progress too. Ask the model to keep notes across sessions, then compare your first and last answer to the same question. Repeat weaknesses jump out, and visible improvement motivates better than any advice.

Interviewing in a second language

If the interview is in English and English is not your first language, AI solves two separate problems. Vocabulary: ask for thirty to forty stock phrases for your role, from small talk to salary conversations. Fluency: the same voice mock interviews, in English, with the instruction to correct only the mistakes that break understanding. The goal is speaking freely, not speaking perfectly.

Real-time AI interview assistants: how they actually work

This is the loudest part of the conversation. A live assistant is software that listens to the call and shows you a prompt while it is happening. Over 2025 and 2026 these tools grew into their own category: Cluely, Interview Coder, Final Round AI and Vizir. Side by side comparisons live on our alternatives page.

What the software does

Take Vizir as the worked example. The app listens on two channels, your microphone and the other person's audio, and transcribes both in real time. When a question lands, a hotkey puts a prompt or a full answer on screen. Technical rounds get their own path: one shortcut grabs the problem statement from your screen, the next one sends it off and writes the solution.

The prompt appears on an overlay, a window layered on top of your screen that stays out of a browser screen share. The word that matters is prompt, not autopilot. The software suggests; you answer, in your voice, in your words, from your experience. Role by role breakdowns, from engineer to product manager, live on the use cases page.

When it genuinely helps

Three situations, mostly. Stress, where things you know cold evaporate the moment you are asked. Recall questions, where you need a date, a number or a detail from a project three years back. And interviews in a second language, where a prompt removes the fear of losing the phrasing mid-sentence. In all three the assistant is not answering for you. It is handing you back access to what you already know. The same thing is true outside hiring, which is why people keep it running on ordinary work calls too.

Where the overlay shows up and where it does not

This is the most important section of the guide. Only you see the hints, and the question is which screen they land on. No assistant can promise you will not be noticed, and any tool that implies otherwise is selling marketing rather than engineering. What is actually true depends on the format:

The full mechanics are laid out on the visibility map. For the formats where the whole screen can be captured, Vizir mirrors hints to a phone companion over an encrypted channel, and your phone is not part of any screen share.

Where it fits and where it does not

• Fits: online interviews where the screen is shared from a browser, screening calls and conversations with no sharing at all, technical rounds on web coding platforms.
• Does not fit: anything with webcam proctoring, in-person interviews, and any format whose rules explicitly ban outside software.

Rules, risk and where you stand

Preparing with AI and rehearsing with it draw far less objection than live prompts, though plenty of employers have their own rules even about those. Live prompts during an interview are a grey area, and grey areas deserve an adult response rather than a wink.

• Company and platform rules. Plenty of employers and assessment platforms explicitly forbid outside software and any assistance during an interview. Breaking that can mean rejection, a withdrawn offer, or a ban from the platform. Read the rules for that specific stage before it starts, not after.
• Proctoring. Webcam systems watch gaze, head movement and your surroundings, and they scan running processes. An overlay does nothing against them, and it is not worth attempting: the price of being wrong is a voided result and a story that follows you.
• The law is not one law. What software may listen to, and what counts as consent, differs by country and sometimes by state or region. Check the rules where you live and the policy of the company you are interviewing with. This article is not legal advice.
• Responsibility. You make the call to use an assistant, and the consequences are yours. No tool absorbs them for you.
• Life after the offer. If prompts covered real gaps in your knowledge, you will be closing those gaps on the job without any prompts. Passing an interview is easier than passing probation.

The reasonable position: a live assistant is insurance against stress, memory blanks and a language barrier. It is not a substitute for competence. It helps you show what you know, not fake what you do not.

Six mistakes people make

1. Reading the prompt word for word. Flat delivery, a fixed stare and the same pause before every answer are obvious to an interviewer without any screen capture involved. Use the prompt as a handrail: skim it, then say it your way.
2. Skipping preparation because the assistant will cover it. A prompt helps you recall and structure what you know. Nothing helps you survive follow-up questions on somebody else's text.
3. Shipping raw model output. Generic phrasing with none of your own numbers or experience sounds identical across candidates, and interviewers have learned to spot it.
4. Not checking the format in advance. Turning an assistant on during a proctored stage or a native screen recording is the worst case. Always find out how the round runs: browser, native client, recording, proctoring.
5. Never rehearsing out loud. Written notes collapse at the first sign of stress. Speaking is only trained by speaking.
6. Memorising your intro verbatim. Hold the structure and the key points, pick the words live. That is what sounds human.

A one week plan before the interview

1. Five to seven days out. Break down the job post with a model, build the question list, write your own answers, run them through the critic prompt.
2. Three to four days out. Assemble your "tell me about yourself", tailor the resume, do two or three voice mock interviews, in your interview language.
3. One day out. Find out how the round actually runs. If it is a normal call or a web coding platform with no proctoring, and you want the safety net, set the assistant up and test it in advance.
4. On the day. Answer for yourself. The AI is in the prompt box; you are the one on stage.

If you want to try live prompts, Vizir has a free trial with no card required, and it runs on macOS and Windows. Plans are on the pricing page, and common questions are answered in the FAQ. Download Vizir and try it on a practice call first: it takes a couple of minutes and it is the only way to see the visibility map on your own machine.