When a Bot Joins Every Meeting, Where Do Your Notes Live?

The backlash against AI meeting notetakers is not really about the bots. It is about the record they leave behind. Once a tool transcribes a call, the question nobody answers is where that text ends up, who can read it, and what you actually own once the meeting is over.

The data now backs that up. Among professionals who have not adopted an AI notetaker, half (50%) cite privacy and security as their main reason for holding back ¹. Even among active users, 47% say a tool has recorded or shared something they did not intend ². And one finding reframes the debate: 84% say they modify what they say when a notetaker is present ³. The bot does not just record the meeting. It changes it.

What people actually believe about meeting bots

Most people treat an AI notetaker as a harmless productivity hack: it joins, it listens, it emails a tidy summary, everyone moves on. The pitch is real time saved, and the convenience is real too. That charitable reading is exactly why three bots can end up silently sitting in one call before anyone in the room stops to question it.

The trouble is that the convenience hides a hand-off. Law firm authors Jenn L. Malik and Peter D. Zittel put the discomfort plainly: "Most companies would never allow an unknown third party to sit in on executive level strategy sessions, legal consultations, or sensitive personnel discussions. Yet AI meeting assistants now perform a functional equivalent of that role, often without formal approval, policy guidance, or executive awareness." ⁴ The bot is a participant most teams never voted to admit.

The anxiety is about the destination, not the AI

The real anxiety underneath the backlash is not "AI is bad." It is "I do not know where this record goes." A transcript is data, and most platforms send it somewhere you cannot see. The bot is less a notetaker than a pipeline, and the destination is rarely your own device.

Malik and Zittel describe the route bluntly: these tools "transmit audio to third-party servers for processing," and "many platforms rely on vague or inconsistent disclosures that do not clearly explain who is recording, how the data will be used, or where it will be stored." ⁵ The transcript lands on someone else's server, under someone else's terms. When 47% of users report unintended capture or sharing ², they are describing a record that traveled further than they meant it to.

The consent gap the lawsuit illustrates

A pending class-action makes the destination problem concrete. In Brewer v. Otter.ai, a participant with no Otter account alleges he was recorded on a call run by another attendee ⁶. The complaint alleges a notetaker "may join the meeting without obtaining the affirmative consent from any" participant, and used recordings "to train its ASR and machine learning models." ⁷

These are allegations in an unproven complaint, filed August 15, 2025 in the Northern District of California and later consolidated as In re Otter.AI Privacy Litigation; no substantive rulings have been issued yet ⁸. Treat them as alleged. The case is useful precisely because it is a documented illustration of the gap people feel: a record can be created, stored, and reused by a third party without the affirmative consent of everyone in the room. That is a destination problem wearing a consent label.

One notetaker, a dozen consent laws

The legal terrain makes the destination question harder. Federal wiretap law and many states follow a one-party consent rule, "but about a dozen states require all participants to consent to being recorded." ⁹ HR Executive corroborates: "approximately a dozen states require all participants to consent to the interception or recording of a conversation." ¹⁰ One bot, many jurisdictions.

For a distributed team that is a live trap. Michael Goldfarb, founder and lead attorney at Guardian HR, frames it well: "If you have a hiring call with a candidate in Illinois, your HR manager in California, and a hiring manager in Florida, you could be dealing with three different legal frameworks at the same time." ¹¹

The consent rules drift, and the list of all-party states is not stable enough to print as current law — "about a dozen" is the honest framing. None of this is legal advice. It is a map of why "just turn on the bot" was never a neutral act.

Capturing a meeting and owning the record are two jobs

The backlash blurs two different jobs. One is capturing the meeting — turning speech into text. The other is owning the record — controlling where that text lives afterward. A visible bot does both at once, which is why the anxiety attaches to the bot itself. But the second job is the one people actually care about.

You can see the conflation in how the market is responding. Privacy is now the dominant adoption barrier: one vendor research report found 73% of businesses cite privacy concerns as the primary barrier to broader adoption ¹². The same discourse has produced a name for the felt cost — "bot fatigue," the sense of multiple recorders quietly accumulating across every call. The fix people reach for is not a better bot. It is a record they keep.

What a record you own looks like

A record you own is just text you hold, not a transcript you rent. The properties that matter are capability-level, and none of them depend on a particular vendor staying in business or honoring a privacy promise. They describe where the text lives and who can reach it, rather than which bot produced it:

• Stored locally on your own device, not on a third-party server you cannot inspect.
• Open Markdown you can move anywhere — readable in any editor, exportable, future-proof.
• Works offline, so the record exists even when nothing is connected.
• End-to-end encrypted when you share it, so the route is closed, not vague.
• No account in the loop, so there is no silo quietly accumulating your meetings.
• Bring-your-own-key AI if you want a model to summarize — you choose the model, rather than handing the full transcript to a vendor by default.

This is the architecture-level answer to the destination question. A meeting summary belongs in a plain Markdown note you hold, not a vendor transcript silo someone else can read or train on.

The thesis here is the meeting-record instance of a broader principle — that you should be able to own your data beyond privacy, and that what you type into AI leaves your walls the moment it hits a hosted model. It is the same problem your notes face when the company behind your app shuts down.

The honest limits of owning the record

Owning the record solves less than it sounds, and saying so is the point of an honest argument. Two caveats are load-bearing, and skipping them would oversell the fix. Ownership changes the destination of the record going forward; it does not rewrite the law of the meeting or claw back what already left.

First, ownership is not consent. Keeping the only copy of a meeting on your own device does nothing to relieve your duty to tell everyone in the room they are being recorded. The all-party-consent map still applies. Ownership is about the destination of the record, not a license to record.

Second, owning a copy does not retract what a hosted bot already holds. If a tool already transcribed and stored a call, exporting a clean copy into a note you control does not pull the original back from the vendor's servers. Ownership matters going forward — for the copy you keep and govern — not retroactively.

The practice: what to do before the next bot joins

The shift is small and immediate, and it costs nothing to start. Before the next meeting, ask one question out loud and change one default. The goal is to make the destination of the record a deliberate decision rather than whatever the bot does by itself:

1. Ask the destination question out loud: "After this call, where does the only record live, and who can read it?" Make it as routine as asking who is taking minutes.
2. Disclose the recorder every time. Whatever the consent law where you sit, treat all-party disclosure as the floor. It is cheaper than a dispute.
3. Decide capture and ownership separately. If a bot must transcribe, plan where the clean copy lands afterward — in a note you hold, in open Markdown, not only in the vendor's silo.
4. Default to a record you control. When the meeting matters, capture it into text on your own device first; reach for a hosted summarizer only once you have chosen what leaves.

Frequently asked questions

Where do AI meeting notes and transcripts actually go?

Most AI notetakers send audio and text to third-party servers for processing. As one legal analysis notes, many platforms "do not clearly explain who is recording, how the data will be used, or where it will be stored." ⁵ The transcript typically lives on the vendor's infrastructure under the vendor's terms — not on your device — unless you deliberately export a copy you control.

Should I let an AI notetaker record my meetings?

That is a consent decision before it is a tool decision. About a dozen states require all participants to consent to recording ⁹, so disclosure is the floor wherever you sit. If you do record, decide separately where the clean copy will live afterward — owning the record is a different question from being allowed to create it.

Do AI notetakers train on my conversations?

Some have been accused of it. The complaint in the Otter litigation alleges recordings were used "to train its ASR and machine learning models" without adequate disclosure ^{7 13}; those claims are unproven and no rulings have issued ⁸. The honest takeaway is to read the disclosures, and to treat "we do not train on your data" as something to confirm in writing, not assume.

Is it legal to record a meeting without everyone's consent?

It depends on jurisdiction. Federal law and many states follow a one-party consent rule, but about a dozen states require all participants to consent ^{9 10}. For a multi-state call you may face several frameworks at once ¹¹. The state lists shift over time, so treat all-party disclosure as the safe default and consult counsel for specifics — this is not legal advice.

Why are people refusing AI notetakers in meetings?

Because the bot changes the conversation and the destination of the record is unclear. 84% of professionals say they modify what they say when a notetaker is present ³, and legal commentators compare the bot to "an unknown third party" admitted to sensitive discussions without approval ⁴. The refusal is less anti-AI than pro-ownership: people want to know where the record goes.

Who can read my AI meeting transcript?

Often more parties than you would expect, and the disclosures rarely say. Analysts note that platforms transmit audio to third-party servers and offer "vague or inconsistent disclosures" about who is recording and where data is stored ⁵. The most reliable way to bound the audience is to keep the authoritative copy in a record you hold and share it under end-to-end encryption.

The question to ask before a bot joins is not which notetaker is best. It is where, after this meeting, the only record of it will live.

If keeping the record on your side is the point, mnmnote.com is a browser-based Markdown editor where notes stay on your own device, work offline, and travel as plain text you can move anywhere.