AI Job Interviews: What They Actually Score, and How to Not Waste Your One Take

You apply for a role that looks like a strong match. A few days later, an email asks you to record an "on-demand interview" by Friday. There is no recruiter on the call, no small talk, just a webcam, a written prompt, and a countdown timer. You answer, click submit, and the recording disappears into a system somewhere. You have no real idea who, or what, is going to watch it back.

That experience has quietly become common. Some first-round interviews now start with software instead of a person: a set of preset questions delivered by text or a recorded voice, your answers captured on camera, and an algorithm doing the first pass before anyone human sees the result. If being interviewed by a machine feels strange, that reaction is normal. This guide covers what an AI interview actually checks, what happens if you'd rather not do one, and how to prepare for it without pretending it's a normal conversation.

What counts as an "AI interview"?

The term covers two fairly different formats. The first is the asynchronous, one-way interview: you get a link, a question appears as text or a recorded clip, and you record your answer alone, on a timer, with no one reacting in real time. The second is a conversational format, where a chatbot or a synthetic voice asks follow-up questions based on what you just said, closer to a live phone screen. Both are different from a standard video interview with an actual human on the other end of the call. In an AI interview, nobody is reacting to you while you talk, which is exactly why it feels so unnatural the first time.

Employers use this format mainly to handle volume. When one posting draws hundreds of applications, a recruiter cannot realistically schedule a live call with everyone. Recording on your own time removes the scheduling problem, and it puts every candidate in front of the same questions under the same conditions, at least in theory.

What the software actually scores, and what it doesn't

Most tools work in a similar way: they transcribe what you say, then score the transcript against the skills and competencies listed for the role. What they weigh most heavily is the content of your answer, meaning whether you mention the experience and language the job description is looking for, and how clearly you structure it. Some also track delivery, like your pace and how many filler words you use.

The part people worry about most is whether the system is reading their face. For one of the largest vendors in this space, that fear is outdated. HireVue said in 2020 that it had removed facial expression analysis from its assessments, after finding that nonverbal signals added very little to how well its models predicted job performance, in some tests as little as a fraction of a percent (Fortune, 2021). That doesn't mean every platform works the same way, but the single biggest name in this space stopped scoring faces years ago. What most tools still lean on is what you actually say.

How common is this, really?

It's more common than most job seekers expect, though the exact scale depends on the survey. In an October 2024 survey of 948 U.S. business leaders, 23% said their company was already using AI to conduct interviews, and of that group, about one in four let AI run the entire interview with no human reviewing it at any stage (ResumeBuilder.com, 2024). The same survey found most companies planned to expand their use of AI in hiring the following year, so the share doing this today is likely higher. In other words, an AI interview isn't a gimmick from one or two startups. It's a real, growing part of the early hiring funnel.

Can you say no, and what happens if you do?

You can decline, and plenty of people do. In practice, refusing a recorded interview usually ends your candidacy for that specific role right away, the same way missing a scheduled call would. That's the honest trade to weigh before you hit decline: it's a real option, but it isn't a free one.

There's a middle path worth trying first: a short, polite reply saying you'd welcome the chance to speak with someone directly, and asking whether a live conversation is possible. Sometimes that gets you a phone screen instead. Sometimes it gets silence, which is itself useful information about how the company treats candidates before you've invested more time.

If the format is genuinely difficult for you, for example because of a disability, that's a different situation from simply disliking the format. In the United States, the Equal Employment Opportunity Commission's 2022 guidance on the Americans with Disabilities Act and AI hiring tools says employers generally need to consider a reasonable accommodation, which can include offering an alternative way to interview, unless doing so would be a genuine hardship. You don't need to name a specific law to ask; saying you need a different format because of a disability is enough to start that conversation.

Where you live also matters. New York City's Local Law 144 has required employers using an automated tool to substantially assist hiring decisions to commission an independent bias audit, publish a summary of it, and give candidates at least ten business days' notice before using the tool, with enforcement in effect since July 2023 (NYC Rules, Local Law 144). Illinois has its own Artificial Intelligence Video Interview Act, in effect since January 1, 2020, which requires employers to tell you AI will analyze your video, explain in general terms how it works, get your consent first, and delete the recording within 30 days if you ask (SHRM / Troutman Sanders summary). Most places have no law like this yet, so read whatever consent screen you're shown before you agree to it.

The good answer works whether a human or a system is scoring it, because it's specific, front-loads the outcome, and uses language a recruiter can act on. The weak one is the kind of answer that sounds fine out loud but gives a transcript nothing to grab onto: no situation, no number, no distinct action.

When the algorithm gets it wrong

Automated screening isn't a neutral, risk-free improvement over human bias, and regulators have already acted on cases that prove it. In 2023, the EEOC settled its first AI-hiring discrimination lawsuit against a tutoring company that had programmed its application software to automatically reject female applicants 55 and older and male applicants 60 and older, a decision that affected more than 200 qualified candidates before any human reviewed it. The company paid $365,000 to resolve the case (EEOC press release, 2023).

A larger, still-unresolved case is testing this in court right now. In Mobley v. Workday, a federal judge granted preliminary certification in May 2025 allowing a nationwide collective action to proceed under the Age Discrimination in Employment Act, covering applicants 40 and older who were screened through Workday's platform since September 2020. The case argues that the screening tool disproportionately filtered out older applicants; Workday has said the claims lack merit, and the ruling is preliminary, not a final finding of discrimination (Holland & Knight, 2025). Whatever the eventual outcome, it's a reminder that a rejection from an automated system isn't automatically a fair or accurate read of your fit for a role.

None of this means you should assume you were treated unfairly every time you're rejected after an AI interview. Most of the time, a rejection at this stage just means the answers didn't hit the specific things the role was screening for. But it does mean the process is a filter, not a verdict, and it's worth remembering that when a rejection email arrives within minutes of you submitting a recording.

How to prepare when you get the link

Treat it like a real interview, because the format punishes people who wing it more than a live conversation does. You usually get one take, a hard time limit, and no chance to recover from a rough start the way you could if a person nodded you along.

Go through the job description first and note the specific skills and responsibilities it repeats. Your resume keyword work applies here too: naturally working the language of the posting into your spoken answers, in your own words, helps a system built around keyword and relevance matching follow what you're saying.

For "tell me about a time" style prompts, structure your answer with the STAR method: situation, task, action, result. Lead with the result if the timer is tight. A clear, front-loaded answer reads well to a person and to a transcript-based scoring model, which is one of the few places where preparing for AI and preparing for a person overlap completely.

Fix your setup before anything else. Light your face from the front rather than sitting with a window behind you, raise the camera to eye level, and test your microphone in a quiet room. None of this is unique to AI interviews, it's the same groundwork covered in our interview preparation guide, but a bad recording is harder to recover from when there's no one to read your nerves and give you a second chance.

If you want a wider view of how automated screening is reshaping hiring beyond just interviews, including what happens earlier in the process when your resume is the thing being scored, our analysis of what millions of job applications reveal about AI hiring digs into the resume side of the same shift.

The bottom line

An AI interview is a screening step, not a verdict. It's built to decide who reaches a person, which means clear, specific, well-structured answers matter more than trying to guess what the algorithm wants to hear. Treat the setup and structure seriously, know that you're allowed to ask for a human alternative, and remember that a fast rejection from a system says something about keyword match at a single moment, not about your ability to do the job.