One-Way AI Video Interviews Aren't Scoring Your Face Anymore — Here's What They Score Instead

What a "One-Way" AI Video Interview Actually Is

The format is mechanically simple, which is part of why it feels so exposing. You get a link, usually with a multi-day window to complete it. Each question appears on screen — as text, sometimes as a pre-recorded clip of a recruiter reading it — followed by a short prep window (often 30 seconds to a couple of minutes) to gather your thoughts before recording starts. The recording itself runs against a hard time cap, typically one to three minutes per answer. Some platforms allow a single retake per question; some allow none at all; a few allow unlimited retakes within the time window. There is no live person adjusting the conversation based on what you say — every candidate answers the same fixed prompts in the same order, then a recruiter (and often, first, an algorithm) reviews the recordings afterward, on their own schedule.

That structure is also, on its own terms, more standardized than most face-to-face interviews manage to be — every candidate is asked the identical question, given the identical time, with no interviewer able to unconsciously go easier on someone they happen to like. The tradeoff is that none of the real-time social feedback a conversation normally provides is available to you while you're answering. You're finding out whether you did well entirely after the fact, from people you can't see.

It's the same black-box feeling candidates run into earlier in the process, too — not being able to see what a system is actually checking for is a recurring theme in modern hiring, not something unique to video interviews.

The Face-Reading Era Mostly Ended in 2021

For several years, the most-cited fear about these systems was facial analysis: software scoring micro-expressions, eye contact, or vocal tone as a proxy for traits like enthusiasm or trustworthiness. HireVue, by far the largest vendor in this category, announced in January 2021 that it was dropping the facial-analysis component of its assessments entirely, following an external algorithmic audit and mounting criticism from researchers, regulators, and advocacy groups who argued the practice was both scientifically shaky and prone to encoding bias invisibly.

That's a real, dated, checkable shift in the industry's largest platform — not a claim that no company anywhere still analyzes facial video in any form. Smaller or specialized vendors may implement things differently, and it's reasonable to ask a recruiter directly what a given platform scores. But the direction of the whole category moved the same way HireVue's did: away from reading your face, and toward reading what you actually said.

Eye contact specifically gets its own, more recent research — two 2024-2025 studies found genuine camera gaze still matters to human reviewers, while software that fakes it for you mostly doesn't, which is worth knowing before you spend money on a gaze-correction filter.

What Gets Scored Instead: Your Transcript, Not Your Face

With facial analysis mostly retired, the evaluation shifted to the part of the recording that was always easiest to defend and easiest to standardize: the words. Your answer is transcribed to text, and both the automated first pass and the human reviewer who eventually watches it are largely evaluating the same things a structured, rubric-scored interview evaluates in any format — whether you actually answered the question asked, whether your answer is organized around a clear point rather than a wandering one, whether it contains the specific, job-relevant detail a generic answer would lack, and whether it stays inside the time you were given instead of running over or trailing off short.

This is good news in a very specific way: it means the thing worth optimizing for is exactly the thing this blog keeps coming back to — a well-structured, specific, complete answer — not a warmer tone of voice or a more relaxed-looking face. It's also why generic advice to "just be yourself and relax" undersells what's actually being measured. A relaxed delivery of a vague, unstructured answer scores the same as a nervous delivery of one.

The same structuring principle applies whether or not a human or an algorithm is doing the scoring — a clear situation-action-result shape is what a transcript-based reviewer is actually looking for, because it's the shape that survives being read as flat text with no tone of voice attached to it.

Why It Feels So Much Worse Than a Real Interview (Even Though You're Probably Not Doing Worse)

There's a real psychological reason a one-way interview feels so much more unsettling than a live one, and it isn't just nerves about being on camera. A normal conversation runs on a constant stream of small feedback signals — a nod, a raised eyebrow, a "right, right," an interviewer leaning in or glancing at their notes — that let you continuously adjust your answer in progress. A one-way recording removes every one of those signals at once. You're not just being watched; you've lost the entire feedback loop you unconsciously use, in any normal conversation, to judge whether you're on track.

A 2023 study published in the International Journal of Selection and Assessment (Kleinlogel et al., "The interviewer is a machine!") tested exactly this feeling against actual outcomes, comparing face-to-face interviews with synchronous and asynchronous video formats. Candidates rated the asynchronous, one-way format as less fair, less personal, and more privacy-invasive than a face-to-face interview — the discomfort is real and consistent, not just in your head. But independent observers rating the actual recorded performances found similar performance across formats, and lower measured stress in the asynchronous interviews specifically, despite candidates reporting the opposite impression. The discomfort of not being able to read the room is a real, well-documented reaction — it just isn't the same thing as evidence that you performed worse.

Candidates felt the one-way format was less fair and more impersonal than face-to-face interviews — but observer ratings showed similar performance, and stress measures were, if anything, lower in the asynchronous format.

“The discomfort of losing every real-time feedback signal is real and well-documented. Research suggests it isn't the same thing as evidence that you actually did worse.”

The Rights Most Candidates Don't Know They Have

One genuinely underreported fact about this format: in Illinois, candidates have specific statutory rights over what happens to these recordings, under the Artificial Intelligence Video Interview Act (820 ILCS 42, in effect since 2020). If an employer uses AI to analyze your video for a position based in Illinois, the law requires them to tell you, before the interview, that AI will be used, explain in general terms how it works, and explain what characteristics it's evaluating — and you have to consent before that analysis can happen. The employer is also restricted in who else they can share your video with, and if you ask them to delete it, they're required to delete every copy, including any held by a third-party vendor, within 30 days.

That's an Illinois-specific law, not a universal right — most candidates outside Illinois don't have an explicit statutory version of it, and it's worth being honest about that limit rather than implying otherwise. But the regulatory direction is spreading, not shrinking: New York City's Local Law 144 requires employers using "automated employment decision tools" — a category that commonly includes AI video-interview scoring systems — for NYC-based hiring to undergo an independent bias audit and publish a summary of the results, alongside giving candidates notice that such a tool is in use. Knowing these rights exist is useful even outside the specific jurisdictions that guarantee them: it's a reasonable, normal question to ask a recruiter directly — what is being evaluated, and can my recording be deleted — whether or not the law in your location requires them to answer.

How to Actually Prepare, Given What It's Really Scoring

Once you know the target is a structured, complete, on-point answer read largely as text rather than a warm-looking face, the useful preparation changes. Use the prep window to sketch two or three concrete points you want to hit, in order — not a script to read, which reads as noticeably unnatural on camera, but a shape to follow. Aim for the middle of whatever time cap you're given rather than the maximum: an answer that visibly runs right up against the limit, or gets cut off mid-sentence, reads as unstructured even when the content was fine. Check the platform's retake policy before you start recording your first answer, so you're not discovering a one-shot-only rule partway through.

The single highest-leverage thing you can do, though, has nothing to do with any individual tip and everything to do with the missing feedback loop described above — since there's no interviewer in the room to tell you in real time whether an answer is landing, the only way to find out is to simulate the exact constraint yourself: record a real answer, alone, under a real timer, before the actual interview — and then watch it back as a stranger reviewing a transcript would, not as the person who just lived through saying it.

That discomfort of "flying blind" while recording is also worth naming for what it is rather than letting it derail you mid-answer — the same freezing-under-pressure research that applies to live interviews applies here too, and it responds to the same fix: enough realistic repetition that the format itself stops being the novel, threatening part.

Key Takeaways

If you remember only a few things, remember these:

• HireVue, the largest vendor in this category, discontinued the facial-analysis component of its scoring in January 2021 after an external audit and mounting bias concerns — the industry's direction moved toward scoring transcribed content, not facial expression.
• What's actually evaluated now is close to what any structured interview evaluates: whether you answered the question, whether the answer is organized around a clear point, and whether it includes specific, job-relevant detail — not tone of voice or how relaxed you look.
• A 2023 study (Kleinlogel et al., International Journal of Selection and Assessment) found candidates rate one-way video interviews as less fair and more impersonal than face-to-face interviews — but observer-rated performance was similar, and measured stress was, if anything, lower in the asynchronous format.
• Illinois's Artificial Intelligence Video Interview Act (820 ILCS 42) gives candidates in Illinois-based roles a right to notice, an explanation of what's evaluated, consent before AI analysis, and deletion of their recording within 30 days of request — a real, checkable right, though not a universal one outside that jurisdiction.
• The best preparation is simulating the exact missing feedback loop yourself: record a real, timed answer alone before the actual interview, and review it as a stranger would, since there's no interviewer available to tell you in the moment whether it's landing.