An AI notetaker has sat in on the meetings of one in three American workers. Most of them were never asked.
That gap — between how normal these tools have become and how rarely anyone consents to them — is quietly changing how people behave at work.
The consent numbers
A July 2026 survey of 500 employed US adults found 33.4% had an AI notetaker or transcription bot in a work meeting. Otter, Fireflies, Zoom''s AI Companion, Microsoft Copilot: all common, all a click away, all easy to switch on without saying anything.
Of the people who''d had one in the room, only 34.7% said they were always asked first. Another 36.5% were asked sometimes. And 25.1% were never asked at all.
So a quarter of workers have been transcribed without being told.
What the bot does to the room
This is the interesting part. An analysis of 159,870 meetings found that when a notetaker joins, women speak 9% more and junior staff talk nearly as much as the managers do.
That''s a real win. Recording seems to blunt the informal dominance senior people exercise in unrecorded rooms — maybe because interrupting becomes visible, maybe because contributions become attributable to whoever made them.
The same analysis caught the other half of the effect: everybody gets more careful. More participation, less candour. People say more words and fewer risky ones.
Whether that trade is worth making depends on what the meeting is for. For a status update, flatter is pure upside. For the conversation where somebody needs to admit a project is going sideways, careful is precisely the wrong setting.
The transcript is evidence
Employment lawyers have watched an entirely new genre of workplace dispute appear, built on AI meeting records. Fortune laid out the mechanics earlier this year: bots left running after the humans assumed the call was over, faithfully transcribing whatever got said next.
Anyone who has hung back after a client dropped off to debrief with a colleague should find that a little chilling. The notetaker doesn''t know the meeting ended. It only knows the call is still open.
And unlike somebody''s memory, a transcript is searchable, exportable and admissible.
A norm worth borrowing
Institutions that have actually thought this through land on the same three rules. Tell participants before you switch it on. Get explicit consent. Let anyone opt out if they don''t want to be recorded.
Three additions from the failure modes above. Say it out loud at the start rather than trusting a small icon nobody looks at. Shut the bot off when the agenda ends, not when the last person leaves. And decide ahead of time which meetings never get recorded — the one-to-one, the hard feedback conversation, the discussion where someone has to say something uncomfortable.
The real question
The efficiency case for these tools is strong and mostly correct. But a meeting isn''t only a machine for producing decisions. It''s also where people float half-formed ideas and admit they aren''t sure.
If the recording improves who speaks while narrowing what they''ll say, the answer isn''t to ban the bot. It''s to be deliberate about which rooms it belongs in — and to ask, every single time, before it starts listening.
Image: Diva Plavalaguna, via Pexels
