Just the camera and the mic. A standard HireVue on-demand interview records webcam video and audio per question, produces a transcript, and captures nothing else on your machine. It runs as a web page, and a web page can only reach the camera and microphone you granted it. Screen recording needs a second, different browser prompt that asks you to pick a screen or a window, and that prompt is not part of the on-demand flow.
The channels in play, and the ones people assume are in play
Most of the anxiety here comes from merging two separate things: a camera pointed at your face, which is real and central, and a monitor watching your machine, which is not part of the product.
| Channel | In play? | What it actually yields |
|---|---|---|
| Webcam video, per answer | Yes, this is the product | A clip per question that a recruiter can watch, rewatch and pass around internally |
| Microphone audio | Yes, recorded with the video | Your answer as spoken, pauses and restarts included |
| Machine transcript | Yes, in normal configurations | Searchable text of what you said. Words are what the platform can index and score, so this is the channel that matters most |
| Timing and takes | Yes, the platform sets them | Thinking time, answer length and retakes are its own settings, so it knows how long you took and how many attempts you used |
| Identity photo or stills | Configuration dependent | Some employers enable a photo step, sometimes with an ID document. Treat it as present only if you are prompted for it |
| Coding assessment editor | Only if you were sent one | A separate browser editor. Editors of that kind typically log pastes and submissions inside their own page, and nothing outside it |
| Desktop or screen recording | No, not in the on-demand flow | Nothing. It needs a share picker you never saw and an indicator you would still be looking at |
| Your other tabs and windows | No | Nothing. A page cannot read another origin's tab, let alone another application's window |
| Keystrokes outside the page | No | Nothing. Key events fire only for the page that has focus |
| Installed or running software | No | Nothing. Web pages cannot enumerate processes. That belongs to installed lockdown software, which this flow never asks you to install |
Why a browser tab cannot quietly film your desktop
This is not a promise from a vendor, it is how browsers are built, and you can verify it while the interview is still open.
- Two unrelated APIs. Camera and microphone come from one browser call, screen capture from an entirely different one. A site granted the first has no path to the second.
- Screen capture cannot be pre-approved. Camera permission can be remembered for a site. Screen capture cannot. Every time, a picker opens, you choose the screen, window or tab yourself, and an indicator then persists for the rest of the session.
- The indicators belong to the operating system. Your camera light and the browser's recording badge are drawn outside the page and cannot be faked by it.
- You can audit it mid-interview. Click the padlock in the address bar. It lists which permissions this origin currently holds. Expect camera and microphone. Anything else would be listed there too.
The on-demand flow is a link, not a download. An employer who wants desktop-level lockdown has to send a separate application that you knowingly install and run. On mobile the boundary is the same, enforced by iOS and Android with their own recording indicators.
The facial expression question is genuinely out of date
For years the loudest criticism of HireVue was that its algorithms scored faces: expressions, micro-expressions, tone of voice. That criticism was well founded at the time and drew a formal complaint in the United States in 2019. In early 2021 the company said it had stopped using facial analysis in its assessments, describing the visual component as adding little predictive value beyond the language of an answer. That was widely reported.
Two things follow, pulling in opposite directions. Automated facial-expression scoring is not something to plan around in 2026. But the camera still records, the clip is still watched by people, and no candidate can audit what runs on a video after it leaves their machine. The removal is the company's own statement, not something you can confirm from your side of the connection.
What you can confirm is the paperwork, because in some places it is legally required to exist.
- Illinois. The Artificial Intelligence Video Interview Act, in force since 1 January 2020, requires an employer using AI to analyse video interviews to notify you beforehand, explain in general how the AI works and what characteristics it uses, obtain consent, limit who else sees the video, and destroy it and all copies within 30 days of your request.
- Maryland. Consent is required before facial recognition technology may be used during an interview.
- New York City. Local Law 144 requires annual bias audits and candidate notice for automated employment decision tools used on city roles.
The consent screen in your invitation is therefore the most authoritative document about what is analysed in your specific interview. It is also the screen everybody clicks past.
Integrity claims you cannot check from your side
Like the rest of the assessment industry, HireVue now markets integrity features: identity verification, checks for scripted or AI-generated answers, detection of manipulated video. Label these accurately in your own head. They are vendor statements about server-side processing, and three structural things stop a candidate confirming any of them.
- The analysis runs after upload. None of it is observable in your browser, so seeing nothing in the page proves nothing either way.
- No vendor publishes thresholds or false-positive rates, and publishing them would weaken the control. The accuracy is unknown to you and, in practice, to the recruiter reading its output.
- The product line has absorbed other assessment vendors, Modern Hire among them, so guides written a few years ago may describe a stack that no longer matches what you were sent.
What is solid: the recording exists, a human can watch it, your words are transcribed and searchable, and the disclosure notice is enforceable in some jurisdictions.
Five minutes before you press record
Read the invitation and the consent screen
They state the question count, the time per answer, whether retakes are allowed, and any AI analysis or retention terms. It is the only document specific to your interview. Everything else, this page included, is generic.
Count the permission prompts
Camera and microphone is the expected set. No screen or window picker means no screen recording. If a picker does appear in a later live round, remember that sharing one window is not the same as sharing a whole display.
Treat the practice question as a rehearsal
Not just a mic test. Sit where you will sit, at the volume you will speak, with the light in front of you rather than behind. Watch it back once. Most people find their framing or audio is worse than they assumed.
Respect the per-question limit, and keep notes above the lens
The limit is shown before you record and is often a couple of minutes. An answer cut off mid-sentence is what the reviewer sees. Six bullet points stuck under the webcam keep your gaze near the camera; a document on a second monitor turns your eyes ninety degrees off-lens.
Where OfferPilot AI is the wrong tool for this
Worth stating plainly, because the query that brings people here often hides a different one. A one-way HireVue recording is close to the worst possible fit for a live copilot, for reasons that have nothing to do with policy.
- A camera is not screen capture. Our overlay is excluded from screen capture at the operating-system level, so it stays out of Zoom, Teams, Meet, OBS and screenshots. A webcam pointed at your face is a different channel and that exclusion does nothing about it, nor about lockdown software, an invigilator in the room, eye tracking, or a phone propped up as a second camera.
- There is no interviewer to listen to. Interview Copilot works from the interviewer's voice arriving through system playback. On-demand gives you a written or pre-recorded prompt, a short countdown, then you talking alone.
- Plenty of people record on a phone. OfferPilot runs on Windows 10 and 11 only. No macOS, no Linux, no mobile, no browser extension.
- Reading is visible. A human watches these clips, and eyes tracking across text read differently from eyes thinking. There is no way around that in a medium built entirely from your face.
The honest use is earlier. HireVue prompts are structured and tend to repeat across a company's funnel, which makes them unusually predictable. The company Q&A generator builds a question bank for a specific firm and role, so you can rehearse the dozen prompts you are most likely to face and have a two-minute version of each ready before the timer starts. For the live human rounds after a passing HireVue, and the coding rounds further down, a copilot has a job to do. Here it does not.
The short version
Your screen is not being recorded. Your face, your voice and your word choice are, along with how long you took and how many attempts you used. Facial-expression scoring was dropped years ago according to the company, the newer integrity claims are unverifiable from where you sit, and the consent screen you were shown is the one document describing your actual interview. Prepare for the transcript, look at the lens, and stop worrying about your desktop.
Related guides
- Does HackerRank Record Your Screen? What Each Test Mode Captures
- Is CodeSignal Proctored? Camera, ID, Screen and Tab Rules
- Which Coding Assessment Platforms Are Proctored? 7 Compared
- What Respondus LockDown Browser Can Actually See, and Cannot
- Is a Karat Interview Recorded? Who Watches the Video and When