AI Interview Screening in 2026 — How These Rounds Run, What They Evaluate, and How to Rehearse

Common enough to prepare for by default. Greenhouse surveyed 2,950 candidates across the US, UK, Germany, Australia, and Ireland and published the results in May 2026: 63% had faced an AI interview, up 13 points from six months earlier. That is the share of job seekers who have encountered one, not the share of employers using them — but from a candidate's side, that is the number that should set your preparation.

The disclosure gap is the practical problem. 70% said they were never clearly told upfront that AI would be evaluating them, and 21% learned only after the interview began; just 18% said the employer had a clear AI policy at all. That is why "I'll prepare differently if it turns out to be an AI round" does not work as a plan.

These rounds also carry consequences. 38% of candidates in the survey said they had walked away from a hiring process because of an AI interview, and 51% said they never heard back after completing one. Whatever you think of the format, it is a screening round with a pass or fail on the other side, not a formality.

Two formats dominate, and they ask different things of you. The first is the one-way recorded interview — the format HireVue popularized as an on-demand interview. You get a fixed list of questions, you record your answers alone in front of a webcam on your own schedule, there is no interviewer reacting to you, and each question comes with a short window to think and a capped window to record. Employers configure those limits, so read the invitation instead of assuming.

The second is a conversational AI interviewer that talks with you in real time. Mercor's candidate documentation describes its version: roughly a 20-minute interview, a working camera and microphone required, role-specific questions, a transcript generated from the session, and up to three retakes across all applications tied to that interview. Mercor also reuses one completed interview across multiple applications, so a single sitting can gate several roles at once.

Both formats usually run in a browser, and both punish setup problems more than a human interview would. Mercor's own guidance tells candidates to use a quiet environment and headphones with an external microphone, and to keep answers brief enough that the AI does not interrupt. A person will wait out a barking dog; a recording timer will not.

Start with an honest boundary. No major vendor publishes the weights inside its scoring model, and any guide that tells you exactly what a specific AI interviewer rewards is guessing. What is on the record is the direction the industry moved — toward what you say rather than how your face looks while you say it.

HireVue is the documented case. It stopped using facial analysis in its assessments in March 2020 and announced the change in January 2021, alongside an algorithmic audit by O'Neil Risk Consulting & Algorithmic Auditing. The company's stated reason was that advances in natural language processing had made language analysis predictive enough that visual analysis no longer added meaningful value. Mercor's documentation likewise describes a transcript of the interview being used to evaluate a candidate for the listing.

The practical read is narrow but useful: your words, in a transcript, are the durable signal. Content that survives transcription — a specific situation, what you personally did, a result someone could check — carries. Rapport, gestures, and the read-the-room adjustments that rescue a human interview do not. That is a preparation instruction, not a claim about anyone's rubric.

It depends on where the hiring happens, and the rules are a state-by-state patchwork rather than one federal standard. Illinois has had the Artificial Intelligence Video Interview Act (820 ILCS 42) in force since January 1, 2020: before an AI-analyzed video interview, an employer must tell you AI may be used, explain how it works and what general types of characteristics it evaluates, and obtain your consent. On request, the employer must delete your interview — and instruct anyone who received copies to delete theirs — within 30 days.

Other jurisdictions cover different pieces. New York City's Local Law 144 requires an automated employment decision tool to have had an independent bias audit within the past year, requires a summary of that audit to be published, and requires candidates to be notified at least 10 business days before the tool is used on them. Maryland's HB 1202 requires written consent and a signed waiver before facial recognition is used in an interview. In Illinois, HB 3773 took effect January 1, 2026, making it unlawful to use AI in employment decisions in a way that discriminates on a protected basis.

Where no statute applies, asking is still normal, and candidates want more of it: in the Greenhouse survey, 57% believed disclosure should be legally required, 44% wanted upfront disclosure, 39% wanted a clear explanation of what the AI measures, and 46% wanted the option to request a human interview. Will AI evaluate this round, and will a human review the result — those are reasonable questions to put to a recruiter.

Rehearse out loud, in one take. The largest gap between candidates who do well in recorded rounds and candidates who do not is that the second group prepared by writing answers down and reading them over. Speaking is a separate skill: the sentence you can write in thirty seconds takes ninety to say, and a written answer read aloud sounds read aloud. Record yourself, play it back, and count how long you actually take.

Build for the transcript. Lead with the answer, then the situation, what you did, and the result, with numbers you could defend. Keep each answer inside the window the invitation specifies — around two minutes is a common cap in one-way formats — and rehearse the second and third layer of your three or four core stories, because a conversational AI interviewer will follow up on whatever you left vague.

Rehearse the setup too, not just the answers. Quiet room, an external or headset microphone, the browser you will actually use, and one full end-to-end dry run before the real invitation lands. Your first AI interview should be something you have already done, not something you are doing live for an employer.

Preterview is a rehearsal tool for exactly this round. Three AI interviewers read your resume, portfolio, and the job post, ask questions by voice, and follow up on what you actually said instead of moving down a list. The session ends in a report: an overall 0-100 score, a five-competency review, delivery analysis covering pace and filler words, and answer-by-answer feedback with a stronger version of each answer. Content and delivery land in one report, because the round you are rehearsing for is judged on both.

The boundary matters as much as the feature list. Preterview is used before a real interview, not during one. It is not an assistant that feeds you answers while an employer is watching, and that is a deliberate choice rather than a missing feature. We also do not claim to know how HireVue, Mercor, or any other vendor scores you internally, and you should be skeptical of anything that says otherwise. What a rehearsal gives you is the format, the pressure, and a feedback loop.

The honest caveats: Preterview is a small, young service, it runs in the browser and works best in Chrome, and interviews run in English or Korean. It is sold per use rather than by subscription — the Light pack is $6.99 for 5 passes, enough for one full mock interview or five portfolio reviews, valid for 90 days, with no auto-renewal — and sign-up includes a free pass worth one portfolio review. If you want a large public question bank or a human interviewer, another category serves you better.