What is an AI interview? How AI job interviews actually work

This is Part 1 of our six-guide series on AI job interviews. For everything in one place, start with the complete guide to AI job interviews.

The short version

• It’s not one technology. An AI interviewer asks the questions. An AI evaluator transcribes, summarizes, scores or ranks your answers. A hybrid does both.
• What gets judged varies. One vendor says its AI-scored video interviews score only what you say;¹ another advertises reading visual cues.⁴
• Where judgment happens matters more than whether a camera is on. A typed chat can still shape who moves on, depending on how the employer uses its output.
• “A person decides” only counts if that person can disagree, with the information, authority and time to overrule the machine.
• Five questions clear up most of the fog. They’re at the end of this guide, and Part 2 shows how to ask them.

Types of AI interview: interviewer, evaluator or hybrid

“AI interview” is a bit like “vehicle”: it covers a scooter and a freight train, and you would prepare differently for each. Underneath the label, the AI does one of two jobs, or both.

• AI interviewer: the AI asks the questions. Simple versions read out a fixed list. Newer ones listen to your answer and choose a follow-up, as Talview says its Ivy interviewer does with “Real-Time Follow-Ups.”⁴
• AI evaluator: the AI works on your answers. It may transcribe, summarize, score or rank them, or flag something for a person to check. A human, or plain software, may be the one asking.
• Hybrid: the AI asks the questions, evaluates the answers and hands a report or recommendation to a recruiter.

This is our own simplified sorting, not a count of what employers use. The first and last rows are the two ends of that scale, not extra kinds. The useful question is not whether a webcam is on but where the consequential judgment happens. HireVue, a hiring-software company, says its AI-scored video interviews are scored only on what you say, even though a camera is recording.¹ And a voice call with no camera at all can be both asked and scored by software.²

What it’s like: text, voice, video or AI notes

Text chat. You type answers to written questions. Sapia, for example, says applicants can finish its chat interview on their phone and describes “five structured questions in an untimed text chat”: “No video, no voice, no face. Text only, by design.”⁵ Its candidate page says five to seven questions, so check your invitation.⁶ No camera, by Sapia’s account, so bad hair days are, for once, off the agenda.

Voice or phone. You speak, and where a system scores your words it first turns your speech into text; HireVue, for one, describes a speech-to-text step.¹ In June 2026 HireVue announced a voice-based AI Interviewer that it says runs “dynamic, two-way voice interviews.”² Without a face to read, silence is harder to interpret, so a short pause can feel long.

Video. This covers two different things. In a one-way, or asynchronous, video interview you record answers on your own time and nobody is on the other end live. In an AI-led video interview an agent asks the questions and may follow up; Talview’s Ivy, mentioned above, is one example.⁴ Either way, it can feel like leaving a very long voicemail for someone important.

A person, with an AI taking notes. Some employers pair a live interviewer with an AI note-taker; Metaview, for example, says its notetaker “appears as a clearly named participant that the host admits and can remove at any time.”⁷ The person asks the questions and the AI produces the record others may read later. Talk to the person, not the software, and if an extra named participant joins the call, it is fine to ask what it records and who reads the notes.

What does an AI interview actually evaluate?

Systems differ most here, and rumors do the most damage. There is no single scoring method; three vendors’ own descriptions, as of October 2026, show the range:

• The words only. HireVue says its AI “relies exclusively on what candidates say” and does not analyze facial expressions, body language, your surroundings or tone of voice.¹
• Visual cues as well. Talview says Ivy is the “Only AI using visual cues to guide interactions & assess” and that it interprets non-verbal cues. Its page does not explain how.⁴
• No camera at all. Sapia’s chat interview is text only, so your typed words are the evidence.⁵

These are vendor claims, not independent findings, and the fine print matters. HireVue’s transcript-only statement describes its AI-scored video interviews; its assessments can also include AI-scored games, and each employer chooses settings such as preparation time and recording attempts, so trust your invitation over a stranger’s forum post.¹ For its voice AI Interviewer, HireVue says responses “can be scored against IO-validated rubrics” (IO is industrial-organizational psychology, the study of people at work), but in the pages we reviewed it did not say which signals count.² When in doubt, ask.

Why transcription quality matters

If a system scores your words, it first has to hear them, and speech-to-text software is not equally accurate for everyone. A 2020 study of five commercial systems found error rates nearly twice as high for Black speakers as for white speakers, and named automated interview evaluation as one possible harm.⁸ Those were 2019-era systems, and the two groups were recorded in different regions, which may explain part of the gap.

A messier transcript does not automatically mean a lower score. A Journal of Applied Psychology study of automatically scored mock video interviews found transcription accuracy differed by language background, race and gender, yet those errors did not translate into meaningful bias in the scores.⁹ That’s one study of mock interviews, not a promise about your system. A decent microphone and a quiet room help; Part 2 has the tech check.

What EU law bans: inferring emotions, as of October 17, 2026

Since February 2, 2025, the EU AI Act has banned AI that infers people’s emotions at work, apart from narrow medical or safety uses. The European Commission’s guidelines, which are not legally binding, say the ban covers candidates in recruitment and emotions inferred from biometric data, such as facial expressions, voice or posture, but not sentiment read from written text.¹⁰

In plain terms, an EU employer should not be using AI to read nerves or enthusiasm from your face or voice. AI that scores what you say is not covered by that ban. Depending on its purpose, it can fall under separate rules for high-risk AI, a category that includes AI tools used in recruitment; those rules are due to apply from December 2, 2027.¹¹ This is general information, not legal advice.

Does the AI decide? Who makes the call

Some vendors stress that people make the call. Sapia says: “Sapia.ai recommends. People decide, always.”⁵ HireVue says candidates who opt out of AI scoring take the same assessment and recruiters rate their answers by hand.¹

Not every vendor says so: Talview’s Ivy page does not describe human review.⁴ Either way, those are product claims. Whether your employer offers an opt-out, and how its recruiters use the scores, is up to the employer.

Human in the loop, or human on the button?

“Human in the loop” sounds reassuring. It means something only if the person has the information, the authority and the time to disagree with the machine. A reviewer who sees a score and a recommendation, with a long queue still to clear before lunch, is closer to an approve button than a decision-maker.

A few laws now spell out what review should mean. As of October 17, 2026:

• Colorado, from January 1, 2027. SB26-189, signed May 14, 2026, covers important decisions, including about employment, made on or after that date. If a covered automated tool then materially influences a decision against a job applicant who lives in Colorado, the applicant will be able to request an opportunity for meaningful human review and reconsideration, to the extent that is commercially reasonable. The law defines that review much like the test above: a trained reviewer with authority to change the decision, who looks at the underlying evidence and does not simply defer to the system.¹⁷ The Attorney General still has rules to write, and press reports say the office will not enforce the law until rulemaking ends, so check its site for the latest.¹⁸
• European Union (GDPR). The GDPR protects you from decisions based solely on automated processing that significantly affect you. Where such a decision rests on a contract or your explicit consent, you can ask for human intervention, give your view and contest it.¹²

Each rule applies only in its own jurisdiction and only to the employers and tools it covers. This is general information, not legal advice. Part 5 goes through what you can ask for, place by place.

A short history of AI interviews

These are dates we could verify, mixing laws, research and vendor announcements: a selection, not a full history of AI interviewing.

That January 2020 video study shows a catch with this kind of analysis. Its model learned to copy what interviewers perceived about people; the study did not test how anyone performed at work, and its authors warned the model “might reproduce human biases.”¹⁹ Teaching a machine to agree with an interviewer’s gut feeling is not the same as teaching it who will do the job well.

The rules differ by place, and some are still changing. Two ideas recur: using software does not exempt an employer from employment law,¹⁶ and several laws expect candidates to be told when AI is involved.¹³¹⁴¹⁷ Each rule applies only where it was made and to the employers and tools it covers, and there is no single national rule in the U.S. This is general information, not legal advice.

Why employers use AI interviews, and how common they are

Vendors pitch scale and consistency: HireVue, for example, says its AI Interviewer runs interviews “with every applicant 24/7.”² For an employer with thousands of applicants and a handful of recruiters, that is a big part of the appeal.

How common are they? No one has an independent count. In 2026 research by SHRM, the Society for Human Resource Management, 27% of organizations were using AI in recruiting, the HR area with the highest adoption, and 39% were using AI somewhere in HR. That covers uses such as résumé screening and interview scheduling, and SHRM gives no figure for AI-run interviews,²⁰ so the figures show that AI in recruiting is far from universal, but not how many employers run AI interviews. Whether you will meet one is worth finding out rather than assuming.

Does it work? One large preregistered field experiment, at one firm hiring for entry-level customer-service jobs, found AI-interviewed applicants got offers slightly more often, with people making every decision;²¹ as of October 2026 it is a working paper, not yet peer reviewed, and the complete guide explains what it does and doesn’t show.

How people react to AI interviews, according to research

If an AI interview feels strange, research suggests you are not alone. One complaint even has a name: “opportunity to perform,” the sense that you got a fair chance to show what you can do. In a German experiment, 160 employees read about hypothetical hiring processes; AI in the phone or video interview stage lowered their sense that they could show their skills and made the process feel creepier. AI that only pre-sorted applications, and was openly disclosed, drew no significant negative reaction.²²

Secrecy made it worse. In a small follow-up, 40 participants read a fictional news report revealing that the company had quietly used AI to rank applicants and analyze their voices, and afterward rated it less fair, less trustworthy and less attractive than before.²² It is one more reason to ask up front rather than find out later.

Less natural is not the same as worse. In the field experiment above, applicants rated the AI interviews as less natural, yet reported similar stress and comfort to those interviewed by people. Given the choice, 78% of those who chose picked the AI interviewer, though those who picked it had lower language and analytical test scores on average.²¹

One caveat: feeling a process was unfair is not proof that it was, and feeling fine is not proof it was fair. The remedy the research points to is the one this series keeps returning to: knowing what the system does and who decides.

Five questions to ask before your AI interview

Most of the uncertainty above disappears once you have answers to these. Ask the recruiter before you start, in writing if you can; Part 2 has a short script.

1. What role does AI play in this interview? Does it ask the questions, score my answers, recommend candidates, or only take notes?
2. What information is actually evaluated? The transcript of what I say, my voice, video, visual cues, test answers, my résumé, identity checks, or something else?
3. Who makes the decision? Is the AI’s output advice to a person, or can it advance or reject me automatically? What does that person see?
4. What happens to my data? Who receives the recording and any scores, how long are they kept, and can I have them corrected or deleted?
5. What alternative exists? Can I ask for an accommodation, a different format, a non-AI route or a human review?

Once you know what you’re facing, Part 3 shows how to make your answers clear and easy to follow, and Part 4 covers the traps, including secret AI help during the interview.