Does HireVue Track Your Eyes? What Is Recorded in 2026

Where HireVue denies it tracks eyes, in its own dated words

The sentence sits in a post on the publisher's blog, published 30 September 2021. It reads: "Although we capture videos for later human review, our artificial intelligence only scores what is said by the candidate, and it does not use any visual analysis (meaning that we do not assess or try to interpret a candidate's facial expressions, body language, emotions, or their background and surroundings)."

The same post explains the withdrawal and its reason. "We stopped using video inputs in new models early in 2020, and going forward we will also eliminate speech inputs." The reason given is unglamorous: "Our internal research showed that these were not adding much additional predictive value to our assessments."

A second document repeats the denial and goes further. HireVue's explainability statement, dated February 2022 in its footer, says the AI "relies only on what is said by the candidate and does not use any video analysis or other audio characteristics (meaning that we do not assess a candidate's facial expressions, body language, their background and surroundings, or tone-of-voice)."

That last phrase closes what the 2021 post left open. Speech inputs were promised for removal "going forward," and the 2022 document lists tone of voice among the things not assessed. Both documents are the publisher's and both carry dates, which is as far as sourcing can go here.

So what is the camera actually for?

Human review. The 2021 wording is explicit that videos are captured "for later human review," so the recording exists to be watched by a recruiter or hiring manager rather than measured by a model.

That is the part the yes or no answer leaves out, and it changes what you should do. Nothing about your face is scored. Everything about your recording can be watched by a person who is deciding whether to advance you, and the publisher's own documentation says a mid band interview is viewed by a recruiter before any decision.

Between the recording and the score sits a transcription step. HireVue's explainability statement names a third party supplier, Rev.ai, for turning speech into text, and says that text is what the models then read. Your words leave the platform as audio and come back as a transcript.

One thing in the same family is worth separating out. HireVue's candidate FAQ states that "Hirevue never uses the video recording of your interview in order to identify you, which is sometimes referred to as 'facial recognition.'" That is a narrower promise than the 2021 post, and the difference between the two is where the confusion lives.

What HireVue tracks in 2026, and what it does not

Ask the question that follows the denial, and the answer has moved on since 2021. HireVue now sells an integrity layer, and its product page for it lists what gets watched.

The page states the disclosure rule in its own words: employers should "explain the identity, environment, audio, screen, and browser checks before the interview begins, so candidates know exactly what to expect." It also describes using "candidate snapshots and automated flags for browser-focus loss or unexpected session changes."

Two limits on that table. Every item is configurable, so the checks running on your interview depend on what your employer switched on, and the page says controls are "configured by role risk." And a flag is a signal rather than a verdict: "a single behavior should not automatically be treated as proof of fraud," with people kept in the decision.

The practical reading is short. Your gaze is not measured. Switching windows, pasting an answer in, or leaving the session mid interview may well be logged.

Two dates, and why both of them are true

Pages on this subject give either 2020 or 2021, and the disagreement is real. HireVue's own post says video inputs stopped in new models "early in 2020." The public announcement of the removal came in January 2021, and the post reiterating it is dated 30 September 2021.

Both hold, because they describe different events. An input can be dropped from new models first and removed from the product afterwards, and a company can announce the change a year after starting it. A page that prints one date without saying which event it marks is wrong half the time.

The dates matter for one practical reason. If you sat a video interview before 2020, visual features may have been part of what was modeled. If you are sitting one in 2026, the published position is that they are not, and it has been published twice.

What none of the documents establishes is the state of play in 2026 itself. The strongest denial available is the February 2022 statement. The 2026 pages describe integrity monitoring rather than scoring, and they do not revisit the question.

Why the question keeps coming back

The rumor survives because the publisher's most visible answer addresses a different thing. The candidate FAQ answers facial recognition in the identification sense, which is about naming a person from their face. Behavioral analysis of expression is a separate claim, and it is the one people are actually asking about.

The FAQ does ask the candidate's version of the question, in a candidate's words: "I get really nervous taking interviews and feel a little awkward even in face-to-face conversations. Will I lose a chance at this job if I don't make enough eye contact or if I blink too much?" The answer steers back to competencies and examples, and says to do what feels natural.

There is also the industry backdrop. Video interviewing platforms did market visual features, and the coverage of those features generalized into a belief that every platform scores your face. The retreat was reported far less than the launch.

So the reliable test for any claim like this one is to find a dated sentence on the vendor's own site. For this question, two exist, and both say the same thing.

Eye contact still matters, to the person who watches

Here the honest answer splits in two. No model measures your gaze. A recruiter watching your recording is a person, and stiffness reads badly to people.

HireVue's tips page goes further than most candidates expect and tells you to stop worrying about it: "And remember: don't worry about making eye contact with the camera. Just be natural!" Its FAQ adds a mechanical tip for anyone who wants to look engaged anyway, which is to move the window showing your own video near the top center of the screen, so that watching yourself points your eyes near the lens.

The behavior that costs you is the one the rumor produces. Candidates who stare at the lens to avoid looking away deliver recitations, and that is the flaw a human reviewer notices in the first fifteen seconds.

Notes are allowed, which settles the other half of the worry. The same tips page says you can keep your notes next to you during the interview, so glancing at a card is a permitted movement rather than a penalty.

Your recording: who holds it, how long, and how to delete it

Retention belongs to the employer. The candidate FAQ states it plainly: "Hirevue stores your personal data based on the retention period (for the amount of time) chosen by the company that you're interviewing with. We automatically purge personal data when that company's retention period expires."

The explainability statement gives the number HireVue recommends to its customers, two years, and adds that it "follows company policy." So there is no general answer to how long your video exists, and any page that gives one is guessing.

Deletion is possible and the route is specific. The FAQ says "you can also request that your personal data be deleted at any point in the process," then names who decides: "Hirevue doesn't decide whether or when to delete your data, the company to which you're applying does." It tells you to email the recruiter you have been in contact with, or to check the employer's application page for instructions.

Two collection limits come with a date attached. The 2022 statement says the company does "never collect sensitive data, like protected health information, financial information and we also do not collect dates of birth." The FAQ says assessments "never look at age, race, gender, or even things on your resume such as GPA or the college you attended."

What the denial does not cover

Four things stay unresolved, and they are worth stating rather than filling.

Whether a human watches before you are cut. The documentation says the final hiring decision is human, and that the lowest band is viewed when candidate volumes are short. First round elimination without a person watching is therefore possible, and the publisher does not rule it out.

Which checks are on for your interview. The integrity features are enabled per employer and per role, and nobody publishes the configuration. The disclosure you receive before starting is the only source on your own sitting.

Whether snapshots are on by default. The product page says image capture happens "during eligible OnDemand hiring experiences," which names neither the default nor what eligible means.

What other vendors do. This page is about one publisher. A different platform in the same round may analyze different things, and the invitation email naming it is the only current authority.

What to do instead of managing your face

1. Set up once, for the human viewer. Light in front of you, a quiet room, and one clear test recording played back with the sound up. Audio is the flaw nobody can ignore.
2. Put your notes beside the laptop, six story titles on a card, since the publisher says that is allowed.
3. Answer in words that carry the content. The transcript is what gets scored, so the sentence naming what happened because of you is the one that has to be said out loud.
4. Read the disclosure screen before you start, because it lists the checks running on your session, and it is specific to your employer.

The format itself, including the bands, the retry settings and the games, is set out in our HireVue round guide, and the banking question set covers that variant. Our free drills train the part that transfers to a recorded answer, which is stating a conclusion first and supporting it in two moves, and how an answer is scored publishes the weights they run on, read from the code that does the scoring. Those weights are ours; HireVue publishes no rubric for candidates and no employer publishes a cutoff. Our assessment simulators cover firm built formats, and none of the six is live yet, which that page says itself. Every claim above is a HireVue document dated in the body, checked 29 September 2026, or is labeled as our own reading.