AI Meeting Assistants: What They Actually Record
An AI notetaker joins your call, listens the whole time, and hands you a tidy summary afterward. What that convenience actually involves — audio capture, voiceprints, transcript storage, and in at least one high-profile case, AI training on your conversations — is worth understanding before you let one into a sensitive meeting.
Most AI notetakers join a call as a visible participant, then record, transcribe, and analyze everything said.
Bot-Based vs. Bot-Free Recording
AI meeting assistants split into two fundamentally different capture methods, and the difference matters for what other people on the call actually know is happening. The first and more common approach sends a visible bot — usually labeled with the tool's name — into the meeting as a participant. Otter, Fireflies, Fathom, and Zoom's own AI Companion all work this way: the bot joins through a calendar integration, appears in the participant list, and records and transcribes in real time. The second approach, used by tools like Granola and Krisp, captures audio locally from the user's own device without adding any visible participant to the call at all.
Neither approach is inherently safer from a data standpoint — both are still recording the conversation — but the visible-bot approach at least gives other participants a chance to notice and object, while local capture can happen with no on-screen indication whatsoever unless the platform itself flags that recording is active.
What Gets Recorded, Specifically
Full Audio
The entire spoken conversation, start to finish, not just excerpts or summaries.
Text Transcript
A searchable, word-for-word record generated from the audio, often the most-referenced output.
Speaker Identification
Who said what, matched to names when the tool can identify participants.
Voiceprints
A biometric profile of a speaker's voice, used to recognize them across future meetings.
Video (Some Tools)
Fathom and similar tools can also record video, enabling short shareable clips.
Screen Content
Some tools, including Zoom AI Companion, can optionally use shared screen content via OCR to improve summary accuracy.
The voiceprint piece is worth pausing on, since it's easy to miss. According to privacy advocates cited by the Associated Press, voiceprints function similarly to a fingerprint — a biometric profile unique to a person's voice, created without explicit consent in many setups, that could theoretically be used well beyond the original meeting's purpose.
The Otter.ai Wiretap Lawsuit
In August 2025, a federal class action — Brewer v. Otter.ai — was filed in the Northern District of California, alleging that Otter's "Notetaker" and "OtterPilot" tools secretly recorded private conversations, including those of people who never signed up for Otter themselves, and used those recordings to help train Otter's AI models without adequate consent. The complaint centers on claims under the federal Wiretap Act and California's Invasion of Privacy Act, which allow substantial statutory damages per violation.
Otter's defense argues its Notetaker isn't an independent third-party eavesdropper but simply an extension of whoever the host is, running a tool the host chose to use. In August 2026, a federal judge rejected Otter's motion to dismiss the core claims, allowing allegations under federal wiretap law, California privacy law, and Illinois's biometric privacy law to proceed toward further litigation. The case doesn't resolve whether Otter actually broke the law, but it confirms the underlying legal questions about consent and biometric data collection are serious enough to go forward — and it's a live example of exactly the risk this article is about, not a hypothetical one.
Where Consent Actually Breaks Down
Most meeting platforms — Zoom, Google Meet, Microsoft Teams — do notify participants when the platform's own recording feature is active. What's murkier is third-party AI notetakers layered on top. According to Thorin Klosowski, senior security and privacy analyst at the Electronic Frontier Foundation, some meeting software doesn't clearly indicate when a third-party notetaker is present at all, and many bots are configured to ask only the meeting host for permission — not every participant.
That creates a real gap: a host can enable a notetaker through their own calendar integration, and it will join and record automatically without ever separately asking the other people on the call. Klosowski's practical advice is straightforward: if you're not sure whether a notetaker is active, ask, and treat getting consent from everyone before a sensitive conversation as the more polite — and legally safer — default.
What Happens to the Data Afterward
Beyond the moment of recording, the bigger open question for most organizations is what happens to a transcript after the meeting ends. HR training and certification provider HRCI's CEO Amy Dufrane has been blunt about this risk, telling the Associated Press: "There are huge risks to the organization on AI notetakers. I don't think companies should use it at all." Confidential personnel information, corporate strategy, trade secrets, and offhand remarks that could later be taken out of context all get converted into permanent, searchable text — a fundamentally different risk profile than a live conversation that simply ends when the call does.
Retention Period
Some vendors retain transcripts and metadata indefinitely by default unless a user actively deletes them.
AI Model Training
Some companies use meeting content to train their own AI models — worth checking explicitly per vendor.
Metadata Persistence
Even when content is deleted, associated metadata can remain stored with the vendor.
Text Is Searchable
Unlike raw audio or video, transcripts are cheap to store and easy to search, per the EFF's Klosowski.
Employment attorney Danielle Kays, a partner at Fisher Phillips, recommends explicitly asking vendors whether they retain recordings, transcripts, or metadata indefinitely, and whether that data is used to train AI models — two separate questions with two separate answers, since a tool can promise it won't train on your data while still retaining the raw transcript indefinitely for other purposes.
How to Use an AI Notetaker Without the Downside
None of this means AI meeting assistants aren't useful — for a huge share of ordinary meetings, the time saved genuinely outweighs the risk. The practical fix is being deliberate about when and how a notetaker runs rather than leaving it on by default for everything.
Announce It
Tell participants verbally that a notetaker is active at the start of the meeting, even if the platform shows a banner.
Check the Training Policy
Look specifically for whether the vendor trains its models on your call content, not just its general privacy language.
Set Retention Limits
Many tools, including Zoom AI Companion, let you configure auto-deletion of summaries after a set time period.
Turn It Off for Sensitive Calls
Personnel matters, legal discussions, and unreleased business information are reasonable categories to record manually or not at all.
Bottom Line
AI meeting assistants record far more than a simple summary implies — full audio, searchable transcripts, speaker identification, and in some cases voiceprints and video, retained on a vendor's servers for a period the user often doesn't control. The Otter.ai litigation shows the consent question here isn't theoretical: courts are actively weighing whether current practices meet wiretap and privacy-law standards. The practical answer isn't to avoid these tools, but to be explicit about consent, know the retention and training policy of whatever tool your team uses, and turn it off deliberately for conversations that shouldn't become permanent, searchable text.
More AI Guides, Written Clearly
Browse our full library of AI tool guides, updated regularly as models and features change.
Common Questions
What exactly do AI meeting assistants record?
Most record full audio (and sometimes video) of a meeting, generate a text transcript, identify speakers, and in many cases create a voiceprint used to recognize that speaker in future meetings.
Is Otter.ai facing a lawsuit over recording meetings?
Yes. A federal class action filed in August 2025 alleges Otter.ai recorded conversations, including those of non-users, without adequate consent, and used the recordings to train its AI models. A judge allowed the core claims to proceed in August 2026.
Do AI notetakers ask everyone on the call for consent?
Not always. Many bots ask only the meeting host for permission, and some join automatically through calendar integrations without a separate consent prompt for other participants.
Can AI meeting assistants use my conversations to train their models?
Some can, depending on the vendor and plan. It's worth checking a tool's privacy policy specifically for whether call content is used for model training versus only for generating that meeting's summary.
Are there AI meeting assistants that don't use a visible bot?
Yes. Tools like Granola and Krisp capture audio locally from the device rather than joining as a visible bot participant, which avoids the "extra attendee" problem but still involves recording the conversation.
What's the safest way to use an AI meeting assistant?
Tell participants at the start of the meeting that a notetaker is active, check the tool's data retention and training policy, and turn it off for sensitive conversations involving personnel matters, legal issues, or unreleased business information.
More from the creator
Explore two other projects from the same creator.