AI Video Interviews: A Practical Guide for Hiring Teams

AI interviews let a small recruiting team assess every applicant. Whether they help or hurt depends on the format you pick and on who reads the score before anyone is rejected.

5 August 2026. 8 minute read. Updated 26 October 2026

Nine hundred applications, four backend roles, two recruiters and thirty minute phone screens. The arithmetic does not work, and that is usually the week a team starts looking at AI video interviews. An AI video interview is a recorded or live video conversation in which software asks the questions, transcribes the answers and produces a score or summary that a recruiter uses to decide who moves forward. The idea is simple. The details decide whether it helps, and this guide is mostly about the details.

Adoption is already well under way. SHRM research on the state of AI in HR in 2026, drawing on 1,908 HR professionals, found that 39 percent of organisations had implemented AI in their HR functions, with recruiting the most common use at 27 percent. Whatever your own view, many of your candidates will already have sat one of these interviews somewhere else.

How do AI interviews work?

The pipeline is short. A question set is defined, either fixed in advance or drawn from the job description. The candidate audio and video are captured, speech recognition produces a transcript, and a model scores that transcript against a rubric. The output is a numeric score, a set of ratings by competency, or a written summary with quoted evidence.

The scoring that holds up is built on language. Louis Hickman and colleagues, in the Journal of Applied Psychology in 2022, found that automated video interview models tended to be dominated by verbal behaviour, with extraversion the exception, and that models trained on interviewer ratings showed stronger evidence of validity than those trained on self reports. If a vendor pitches you on reading eye contact or micro expressions, ask what evidence sits behind it, and read the section on the law below before you go further.

What are the three formats of AI video interview?

Three quite different things are sold under one label, and mixing them up causes most poor buying decisions.

In a live AI assisted interview, a human interviewer runs the conversation while the software transcribes, drafts the scorecard and flags any competency the interviewer skipped. Nobody is judged by a machine. The recruiter is simply spared the note taking.

In a one way recorded interview with AI scoring, the candidate gets a link, records answers to fixed prompts alone, and software scores the result. No person is present unless the candidate moves forward. Our guide to one way video interviews covers the settings for this format in detail.

In a conversational AI interview, the software holds a real time voice or video conversation, asks follow ups based on what the candidate just said and probes when an answer is thin. It feels closest to a real interview, and it has the most moving parts.

Format	Person present	Best suited to	Where it struggles
Live AI assisted	Yes	Final rounds, senior roles, anything involving negotiation	Volume, because it saves note taking and leaves calendar hours unchanged
One way recorded with AI scoring	No	High volume fresher and campus screening with clear right answers	Anything that needs a follow up question, and candidates who freeze on camera
Conversational AI	No	Mid volume roles where probing changes the verdict	Highly specialised areas where a partial answer is hard to judge

The pattern is consistent. The more the software decides, the more recruiter time it saves and the more care it needs elsewhere in the process.

Where does AI evaluation genuinely help?

Consistency is the main gain, and it is often undersold. A human panel tends to ask the fourth candidate of the day a shorter, looser version of the question it asked the first. Software asks every applicant the same prompts in the same order, which is the basic condition for comparing people fairly at all.

Coverage is the second gain. In fresher and campus hiring, a single drive can bring in more applications than any recruiting team could phone screen. An AI round often makes the difference between assessing everyone and assessing whoever applied in the first few hours.

The third gain is the least discussed. Every AI interview leaves a searchable transcript of what was actually said. When a hiring manager rejects someone for weak communication, anyone on the panel can go and read the answer. That record improves hiring debates more than any single score does.

Where does AI interview scoring mislead?

AI does not remove bias from hiring. It moves bias into places that are harder to see, and the first such place is the transcript. Allison Koenecke and colleagues, in the Proceedings of the National Academy of Sciences in 2020, tested five commercial speech recognition systems and found an average word error rate of 0.35 for Black speakers against 0.19 for white speakers, a gap they traced mainly to the acoustic models. A score built on a less accurate transcript is a score built on weaker evidence.

A related problem matters in any market where people interview in a second language. Scoring fluency as a stand in for competence penalises the engineer who thinks in one language and answers in another, and rewards the confident generalist who says little at speed. A rubric built around clarity of communication will do this steadily unless you deliberately weight it down for roles where spoken English is not the job.

Then there is coaching and gaming. Some candidates now use AI help during recorded rounds, and the usual sign is a neatly structured answer with no specifics in it. Loose rubrics reward exactly that shape. Follow up questions that ask for a number, a date or the name of the system involved tend to separate memory from recital.

Finally, watch for false precision. A score of 72 out of 100 looks like a measurement, but it is a model estimate compressed into one number. Use it to sort a review queue. Keep it away from senior and leadership hires as a deciding input, because at that level you are judging how someone handles ambiguity, and no rubric captures that well. Our guide on how AI scores an interview goes further into this.

What does an AI video interview cost in candidate goodwill?

More than most teams plan for. Candidates still prefer a person. A poll of 2,019 US adults by the American Staffing Association, reported by SHRM in 2023, found 70 percent preferred in person interviews, 17 percent preferred video calls and 9 percent preferred the phone.

The candidates you lose are not a random sample. People with three other live processes walk away first, which means a poorly run AI round quietly filters for applicants with fewer options. In practice the losses come from the same few gaps, namely no clear notice that AI is involved, no feedback after the interview and no way to reach a person.

Each gap has a simple fix. Say in the invitation that AI will help assess the interview and what it looks at. Offer a human interview to anyone who asks. Have a person review before any rejection, and send every candidate who completes the round a short note on the outcome. In practice these four steps head off most complaints, and none of them costs much beyond a little recruiter time.

What does the law require of AI interviews in 2026?

For hiring into the European Union, Annex III of the AI Act lists AI systems used for recruitment or selection, including those that filter applications and evaluate candidates, as high risk. On 29 June 2026 the Council of the European Union gave final approval to changes that set 2 December 2027 as the start date for stand alone high risk systems. The substantive work of risk management, logging, human oversight and documentation has moved later. It has not gone away.

Some duties need no wait. Article 50(1) requires that people be told when they are interacting directly with an AI system, unless that is obvious. Article 5(1)(f) prohibits AI systems that infer the emotions of a person in the workplace or in education, except for medical or safety reasons, which rules out any interview feature claiming to read stress or enthusiasm from a face.

Data protection law applies alongside. Under the GDPR, Article 22 gives people the right not to be subject to a decision based solely on automated processing with legal or similarly significant effects, and UK guidance names online recruiting with no human involvement as an example. In India, the Digital Personal Data Protection Rules, 2025, notified in November 2025, give candidates rights to access, correct and erase their data, with requests to be answered within a maximum of ninety days. Most employers hiring across borders settle on one policy built to the strictest of these.

Does your hiring volume justify an AI interview?

Count screening interviews per hire. Below about fifteen screened applicants per hire, setup effort and candidate goodwill usually cost more than the round saves.
Check where the bottleneck sits. If offers stall at the final round, an AI screen at the top of the funnel fixes nothing.
Pick roles where a good answer can be described in advance, such as fresher engineering, support, sales development and field roles. Design leadership is a poor fit.
Price the goodwill. If your employer brand is still being built, a recorded and scored round with no human contact can cost you more than it shows.
Decide who reviews before any rejection. If the answer is nobody, stop and fix that first.

Two settings decide whether the round works in practice. Cap it at five questions with answers of around ninety seconds, because few reviewers watch the fourth minute of anything. And allow at least one retake per question, since many candidates record on a phone during a lunch break while still serving notice, and a first nervous take tells you little about them. For a deeper look at the tradeoffs between formats, see live versus recorded interviews.

Sources

02 Questions

Questions this guide answers.

What is an AI video interview?

It is a recorded or live video interview in which software asks questions, transcribes the answers and produces a score or summary. A recruiter uses that output to decide who moves forward. The three common formats are live AI assisted, one way recorded with AI scoring, and conversational AI.

How do AI interviews score candidates?

Most score a machine transcript of the answers against a rubric. Research in the Journal of Applied Psychology in 2022 found automated interview models tended to be dominated by verbal behaviour. The score is only as good as the transcript and the rubric behind it.

Are AI video interviews biased?

They can be, mainly through the transcript and through rubrics that reward fluency. A 2020 study in the Proceedings of the National Academy of Sciences found word error rates of 0.35 for Black speakers against 0.19 for white speakers across five commercial speech systems. Test transcription on your own applicant mix and keep a person reviewing before any rejection.

Do candidates like AI video interviews?

Many prefer a person. A 2023 poll reported by SHRM found 70 percent of US adults preferred in person interviews. Clear notice that AI is involved, an option to request a human interview and feedback afterwards all reduce the goodwill cost.

What does the EU AI Act say about AI interviews?

Recruitment and selection systems are high risk under Annex III, with duties for stand alone systems starting on 2 December 2027 after the Council approved changes in June 2026. Article 50(1) requires telling people when they are interacting with an AI system. Article 5(1)(f) prohibits inferring emotions in the workplace except for medical or safety reasons.

When is an AI video interview worth using?

It is worth using when you screen many applicants per hire for roles where a good answer can be described in advance, such as fresher, support and sales development roles. It is a poor fit for senior or leadership hiring. Always decide who reviews scores before any rejection.

03 Next step

Spend less.Hire best.

Four hiring modules. Free trial with one real candidate.

Talk to us