Job search · Updated 18 September 2026

How to Prepare for a Live AI Job Interview

Nearly two thirds of job seekers say they have now been interviewed by an AI, and most were never told it was coming. Here is what to ask before you accept the slot, and how to rehearse for a conversation that asks follow-ups and gives nothing back.

Prepare for a live AI interview the way you would prepare for a structured interview with a very literal stranger. It asks everyone the same core questions, it asks real follow-ups as you speak, and it judges what your words actually say rather than how the room felt — so depth on four or five genuine examples beats a script, and saying the number out loud beats implying it. The step almost nobody takes is the one with the most leverage: before you accept the slot, email the recruiter and ask whether AI is used in this interview, what it evaluates, and whether a person reviews the result.

This is the first round now, not an experiment

Greenhouse surveyed 2,950 active job seekers and reported in May 2026 that 63% have faced an AI interview — up 13 percentage points in six months. The format varies: some are faceless voice agents on a phone call, some are avatars that smile, blink and nod, in what CNBC's September 2026 write-up called a jarring first contact with a company.

Recruiting consultant Katrina Kibben put the confusion precisely in that piece: telling someone you had an interview "always meant there was a person on the other side that you were interacting with," and now "people are wondering: did the rules change?" Mostly they did not. Almost everything below is ordinary interview preparation, aimed slightly differently.

The largest test of this found candidates did better, not worse

The instinct is that a bot round is a filter designed to thin you out. The biggest study of it found the opposite. Brian Jabarian and Luca Henkel ran a natural field experiment across 70,884 applications for 48 entry-level customer service postings in the Philippines, randomly assigning applicants to be interviewed by a human recruiter or by an AI voice agent. Human recruiters made every hiring decision either way.

Applicants who got the AI interview were offered jobs at 9.73% against 8.70% — about 12% more likely — were 18% more likely to actually start, and were still there at similar margins one, two, three and four months later. On the job, the researchers found "neither statistically significant nor economically meaningful differences" in handle time, customer satisfaction or quality scores between the two groups. And when applicants were allowed to pick, 78% chose the AI voice agent.

The authors credit structure: the agent ran a more consistent interview while still responding to what each person said, so more hiring-relevant information came out. Two caveats, plainly. This was one company's high-volume entry-level pipeline in one country, so it does not transfer automatically to a graduate scheme or a specialist role, and it says nothing about whether the tool in front of you is fair. What it does retire is the assumption that the format is stacked against you by default.

The documented failure is disclosure, not the algorithm

The same Greenhouse survey found where things actually break. Seventy percent of candidates say AI was not clearly disclosed before their most recent AI interview, and 21% only found out once it had started. Thirty-eight percent have walked away from a process because it included one, and 57% think disclosure should be legally required.

The complaint in that data is less about being judged by a machine than about being handed the format with no warning. Walking away is a legitimate choice and plenty of people make it. If you want the job, one email before you accept the slot removes most of the surprise.

What you can ask for, and where it is actually a right

Some of this is owed to you rather than granted as a favour, depending on where the role is based. Illinois amended its Human Rights Act effective 1 January 2026 to require employers to give notice when AI is used in employment decisions, including recruitment and hiring; the state has not yet settled the implementing rules, but the statutory duty is live. If the interview is a recorded video that AI analyses, the older Illinois Artificial Intelligence Video Interview Act also requires the employer to explain "how the artificial intelligence works and what general types of characteristics it uses to evaluate applicants," to get your consent first, and to delete the recording and its backups within 30 days if you ask.

New York City is the interesting one, because it is generous and limited in the same breath. Under the Local Law 144 rules, an employer must give at least 10 business days' notice before using an automated employment decision tool, and the notice "must include instructions for how to request an alternative selection process or accommodation." But the same rules say: "Nothing in this subchapter requires an employer or employment agency to provide an alternative selection process." You are guaranteed the instructions. You are not guaranteed the alternative.

The route that does carry a duty is disability accommodation. The EEOC's guidance for applicants is blunt about how low the bar is: "You must inform the employer that you need some sort of change or adjustment to the application/interviewing process because of your medical condition. You can make this request orally or in writing." You do not have to say the words "reasonable accommodation" or mention the ADA, and you should ask as soon as you know you will need it rather than after the interview goes badly. A timed, silent, no-feedback format is exactly the kind of thing an adjustment exists for. Our guide to recorded one-way video interviews carries the fuller table of what employers owe you by jurisdiction.

It is reading the transcript

Students mostly worry about the camera judging their face. HireVue, the most visible vendor in early-career hiring, removed facial-expression analysis from its assessments — discontinued in March 2020 and announced in January 2021 — saying that language analysis had improved to the point where "visual analysis no longer significantly added value." That is one vendor, so treat it as a strong hint rather than a guarantee.

Either way the practical consequence is the same: say it, do not show it. A shrug, a wince and a knowing pause do not appear in a transcript. Name the team size, the timeframe, the number, the outcome. Habits that feel like over-explaining to a human reader are right here.

Live follow-ups are what break a script

This is the real difference from a recorded screen, and the reason a memorised answer is riskier here than it used to be. A one-way recording asks its question and moves on. A live agent hears what you just said and asks about that — which is the responsiveness the field experiment credits for its results, and which a prepared paragraph cannot survive past the second question.

So prepare fewer examples, deeper. Four or five real situations you can go three layers into — what exactly you did as opposed to what the team did, what the hardest trade-off was, what you would do differently — will outperform ten stories you can only tell once. And rehearse into silence: there is no encouraging nod coming, and the first time you notice that should not be during the interview.

Split the job

The workflow, step by step

Four prompts for ChatGPT, Claude, Gemini or whatever you already use. None of them writes an answer for you.

1. The email you send before you accept the slot. Neutral, short, and not an accusation.

Draft a short, friendly email to a recruiter who has invited me
to a first-round interview.

Role: [title, company]
What I was told about the format: [paste the invitation]
Adjustment I need, if any: [leave blank if none]

The email must ask, politely and in no more than five sentences:
1. Whether any part of the interview is conducted or scored by AI.
2. What the interview evaluates.
3. Whether a person reviews the outcome before a decision.
4. How to request an adjustment to the format, if I need one.

Rules: no accusation, no mention of laws or rights, no implication
that I object to AI. I am asking so I can prepare. Do not invent
anything about the company or about me.

If the answer is that AI is involved, you have just turned an ambush into a known format with a week to prepare for it. If the recruiter cannot say what it evaluates, that is worth knowing too.

2. Rehearse against a deliberately flat interviewer. Use voice mode if you have it — typing is not the skill being tested.

You are conducting a structured first-round interview for this
role:

[paste the job posting]

Rules for you:
- Ask one question at a time and wait for my full answer.
- After each answer, ask two follow-up questions about the
  specific thing I just said - not a new topic.
- Give me NO encouragement, no "great answer", no reassurance,
  no feedback of any kind until the very end.
- Keep a neutral tone throughout. Do not react.
- Stop me at 2 minutes if I am still talking.

Ask 5 questions total. Only at the end, tell me: which answers
had no concrete example in them, and which follow-ups I could
not answer.

The flatness is the point. The discomfort candidates report is largely the absence of a nodding face, and it is fixable by having met it three times already.

3. Find where your story runs out. Depth is the thing a live agent tests and a recording never did.

Here is one experience I plan to talk about:

[paste your 90-second version]

Do not rewrite it and do not tell me whether it is good.

Instead, write the three layers of follow-up a structured
interviewer would ask:
- Layer 1: what specifically I did, as opposed to the team.
- Layer 2: the trade-off or constraint I have not mentioned.
- Layer 3: what I would do differently and why.

Give me the questions only. No answers, no hints, no example
responses.

Answer them out loud. Any layer you cannot reach is a story that will collapse on the second follow-up, and it is better to find that now.

4. Check that your answer survives as text. Record yourself, transcribe it, then hand over the transcript.

Below is a transcript of me answering an interview question out
loud. Read it as someone who cannot hear my voice, see my face,
or know anything about me.

[paste the transcript]

Tell me only:
1. Every point where I refer to something without naming it
   ("that project", "the tool we used", "quite a big team").
2. Every claim with no number, timeframe or outcome attached.
3. What I said in the first 15 seconds, quoted exactly.

Do not rewrite my answer and do not suggest wording. List the
gaps only.

The first 15 seconds matter because each answer is scored against the question that was asked, and an answer that spends its opening warming up has spent its best moment on nothing.

Where the line is

Preparing with AI is ordinary and expected. Being fed by it during the conversation is the one move to avoid — and not only because many employers' terms prohibit it. Our guide to what employers actually allow covers how the policies differ and how to find the one that applies to you.

Related reading

• The other interview formats. Recorded one-way video interviews, behavioural questions and the STAR method, and technical coding interviews.
• Getting to the interview. Passing AI resume screening, tailoring a resume and cover letter, and a LinkedIn profile that AI search can read.
• After the interview. Negotiating a first offer — use AI for the words, never for the number.
• When the recruiter is not real. Spotting AI job scams and fake recruiters, where the money-direction rule matters more than anything about the message.
• Before you paste anything in. The privacy checklist applies just as well to a job description and your own transcript.

FAQ

Is an AI interview a real interview or just a screening test?

Usually a first round that decides whether you reach a human. In the largest study of the format, a 70,884-application field experiment in the Philippines, human recruiters made every hiring decision regardless of who ran the interview. Treat it as a real interview whose output is a transcript that a person will later read.

Can I ask for a human interview instead?

You can always ask, and it is worth asking. Whether anyone must say yes is narrower than people assume. New York City rules require the notice to include instructions for requesting an alternative selection process, while also stating that nothing in them requires an employer to provide one. The claim that does carry a legal duty is a disability accommodation: tell the employer you need a change to the interview process because of a medical condition, orally or in writing, as early as you can.

Is the AI judging my face, accent or background?

Depends on the vendor, and the trend has been away from it. HireVue discontinued facial-expression analysis in March 2020, saying that improvements in language analysis meant visual analysis no longer added significant value. Other platforms are built differently, so ask the recruiter what the tool evaluates. Either way, the safest assumption is that the words are what count, so name things explicitly instead of implying them.

Should I walk away from a process that uses an AI interview?

That is your call, and 38% of surveyed job seekers have done exactly that. Worth weighing against the evidence: in the one large randomised test, applicants interviewed by an AI agent were about 12% more likely to be offered the job and 18% more likely to start it, with no measurable difference in how the hires then performed. That was entry-level, high-volume hiring in one country, so it is not a promise. It is a reason not to assume the format is against you.

Can I use ChatGPT during the interview to help me answer?

No. Beyond the employer policies that prohibit it, it does not work in this format: the agent asks follow-ups about the specific thing you just said, a few seconds after you said it, and the next round is with people who will ask again. Use AI to rehearse, to generate likely follow-ups and to audit your own transcript. Answer live on your own.

Bottom line

An AI interviewer is a structured interview with the small talk removed. The evidence says the format itself is not the thing working against you; not knowing it was coming is. So send the email, ask what it evaluates and whether a person reviews the result, request an adjustment if you need one, and then rehearse the way the format actually behaves — fewer stories, three layers deep, spoken out loud into a room that does not nod back.