Is an AI Notetaker Recording Your Job Interview?

Probably, and the notice you get depends entirely on which tool the employer used. In a Google Meet interview, every participant sees a pencil icon and an explicit announcement. In a Zoom meeting, the only signal may be a small diamond icon changing color. In a Microsoft Teams meeting, Microsoft's own admin documentation does not describe a participant notification at all. And if the employer's applicant tracking system sent a bot, it appears in the participant list as a named attendee that most candidates read as a second interviewer.

This article is about the AI on the interviewer's side of a live video call. That is a different mechanism from proctoring software watching you during an automated assessment, and a different one again from a recorded one-way interview platform. It covers what each of the major meeting tools captures, what notice it gives, whether you can decline, and what actually happens to the transcript afterward. Facts are current as of 2 September 2026. This is a reading of vendor documentation and public employer policies, not legal advice.

Three different things get called "the AI in my interview"

Candidates conflate these constantly, and they behave differently.

Proctoring telemetry runs during an automated coding assessment. It watches your browser, editor, or webcam for integrity signals. It is not a conversation transcript. We covered which platforms actually open a microphone in what coding assessments record and what only sounds like audio monitoring, which deliberately left the live-call case open. This article is that case.

One-way and platform-run interviews — HireVue, Karat — record you inside the vendor's own product, under that vendor's terms, and score the result. The capture is the product. What HireVue records and how it scores covers that surface.

Meeting-side AI notetaking is the subject here. You are on a normal Zoom, Teams, or Meet call with a human interviewer, and something is transcribing it. That something is either built into the meeting platform or joined as a separate bot participant.

What each meeting platform captures, and what it tells you

Google Meet: explicit notice

Google's "Take notes for me" documentation is the clearest of the three. Google states: "Once you turn on this feature, let everyone know you're using 'Take notes for me.' Google Meet informs all meeting participants that notes are being taken and the pencil icon appears on all participants' screens" (Google Meet Help).

The output is an AI-organized notes document, not a verbatim transcript — Meet's transcript capability is a separate feature. Screenshots of presented content are captured if something is displayed for at least ten seconds, which matters if you share your screen for a live coding exercise. Google states the notes document "is saved in the meeting organizer's Google Drive in the 'Google Meet' folder." You are in it; you do not get a copy.

Zoom: a color change on an icon

Zoom's Meeting Summary with AI Companion can be started only by the host or co-host. On notification, Zoom's support article says: "Participants will see the diamond change color as indication that AI features (including Meeting Summary) are active" (Zoom Support).

That is the notice. A candidate concentrating on a system design question is not monitoring the color of the AI Companion icon in the toolbar. Zoom documents that hosts can generate both a summary and an optional transcript, and that summaries are distributed by email and posted to the meeting's group chat — but that "Participants must be signed in to their Zoom account during the meeting to receive the summary automatically." External candidates joining by link are normally not signed in to the host's account context, so the summary of your own interview is generated, distributed, and read without you.

Microsoft Teams: depends on a setting you cannot see

Teams splits Copilot into two modes, and which one is active is an organizer and admin choice. Under "during and after the meeting," Microsoft's documentation states that "Once someone starts transcription, licensed users can select the Copilot button for use during, and after the meeting or event." Under "only during the meeting," Copilot "relies on speech-to-text audio processing data that isn't saved after the meeting or event ends" (Microsoft Learn).

Two things follow. First, the transient mode is genuinely more limited — but Microsoft attaches a caveat: "Depending on your organization's Microsoft Purview retention policies, Copilot prompts and responses during meetings might be retained for compliance purposes, even if recording and transcription are turned off." Second, that admin page does not describe a participant-facing notification for Copilot use. Teams surfaces a banner when recording or transcription is on; Copilot's own presence in the transient mode is not documented there as a separate notice to attendees.

The cross-platform case: Zoom's assistant in a Teams or Meet call

Zoom now lets licensed users send AI Companion into third-party meetings. Its behavior here is the most transparent of anything in this article: Zoom documents that the assistant appears as a visible participant tile carrying Zoom branding, the name of the assistant's owner, and an activity status reading "Transcribing," and that "A chat message is posted 3 minutes after the meeting starts to inform meeting participants of AI Companion's presence and purpose" (Zoom Support).

If you see an unexplained tile or a fourth attendee in a three-person interview, that is what it is.

The bot in the participant list is often the ATS

Recruiting platforms attach their own notetakers. Ashby's documentation describes an AI Notetaker that produces "recordings and transcripts," transcribes in over 99 languages, and joins as a removable participant — Ashby notes that "each video conferencing tool lets you remove the ai notetaker participant from the call natively" (Ashby Knowledge Base).

The consent posture here moved in 2026. On 18 June 2026 Ashby shipped an opt-in consent policy, adding an "Interview Recording Consent" question to application forms and a consent prompt at scheduling, on the stated ground that "many organizations hire in jurisdictions that require affirmative consent, where candidates must explicitly opt in before recording can take place" (Ashby product update). Notably, if a candidate opts in only after the interview is scheduled, "recording remains disabled for that interview" unless an organizer turns it on manually.

You can usually say no, and employers document that it costs you nothing

This is the part candidates most often get wrong, because declining feels adversarial. Published employer policy says otherwise.

eBay's candidate-facing FAQ for BrightHire states that interviews are "recorded and transcribed unless you choose to opt out," that a candidate can opt out before the interview, in the invitation response, at the start, or mid-conversation, and — the material sentence — "Opting out will not adversely affect your candidacy or how you are evaluated. If you choose not to be recorded, interviewers will take notes manually instead." eBay lists retention as "up to 30 days after your interview, unless a different retention period is required by law" (eBay BrightHire FAQ).

Sourcegraph's public recruiting handbook is equally direct: "Using BrightHire is optional for both our interviewers AND our candidates. We will only record interviews with a candidate's explicit consent and they can choose to opt out at any time," and "choosing not to be recorded will have no impact on your candidacy for the role" (Sourcegraph handbook).

Two employers is not a survey, and neither is binding on the company interviewing you. But it establishes that a documented, no-penalty opt-out is a normal thing for a serious hiring team to offer, which makes asking a reasonable question rather than a red flag. What you cannot assume is that the offer is automatic, or that a manager improvising with a personal notetaker subscription has read any of it.

Why this got sharper in mid-2026

The notice-and-consent question stopped being theoretical. HR Executive reported on 5 August 2026 on a suit filed 30 July 2026 in the Northern District of California, Chamberlain v. Granola, brought under California's Invasion of Privacy Act, alleging the product records conversations without telling most participants and uses recordings for model training by default; a consolidated case against Otter.ai in the same court alleges recording without all-party consent and use of that audio for training (HR Executive). These are allegations in pending litigation, not findings.

The reporting flags a detail worth carrying into your own interviews regardless of how those cases end: model training is frequently opt-out rather than opt-in on notetaker plans, and an opt-out typically applies only going forward. The disclosure obligations an employer owes you are a separate question, covered in what AI hiring rules actually entitle candidates to know in 2026.

What to actually do before and during the call

Keep it short and unembarrassed. This works:

1. Ask in writing when you schedule. "Will this interview be recorded or transcribed by an AI tool, and if so, which one and how long is it retained?" A recruiter who cannot answer is telling you something useful.
2. Read the participant list before you speak. A named attendee you were not introduced to, a tile with a "Transcribing" status, or a chat message a few minutes in are all the bot announcing itself.
3. Check the toolbar once. The pencil icon in Meet, the AI Companion diamond in Zoom, the recording or transcription banner in Teams.
4. If you want to decline, do it at the top. "I'd prefer not to be recorded — are you able to take notes instead?" Both employer policies quoted above anticipate exactly this request.
5. Assume the summary outlives the call. Notes land in the organizer's Drive or inbox. Ask for retention in writing if the interview covered anything you would not want stored for months.

What this does not tell you

Vendor documentation describes intended behavior, not what a specific employer configured. Admin settings, retention policies, and integrations vary per tenant, and none of the pages cited here bind the company you are talking to. Recording law also varies by jurisdiction and by where each participant is sitting, which is why the same interview can be lawful for the interviewer and contested by the candidate — a question for a lawyer, not a blog.

There is also a symmetry worth naming. If you are evaluating an AI assistant for your own side of the interview, the questions you should ask a recruiter about their notetaker are the same ones to ask a vendor about yours: what is captured by default, what leaves your machine, how long it is retained, and whether it trains a model. The AI interview assistant privacy checklist works through that in order. Control is a desktop application for Windows and macOS with a free allowance of five messages and two minutes of voice for new users, which is enough to see what it captures before an interview depends on it — check it against that same list rather than taking any vendor's summary, ours included.

The practical takeaway is narrow: in a live interview in September 2026, assume something is transcribing until you have checked, know the three signals that reveal it, and remember that asking for a manual-notes interview is a request serious employers have already written a policy for.