Picture two versions of the same meeting. In the first, six people are on a call with no recording — just a shared doc someone will half-fill in afterward. In the second, everything is identical except a small tile in the corner says a notetaker is listening. Same agenda, same six people, same seniority mix. According to a 2026 study from Read AI, those two meetings do not sound the same. In the second one, women get roughly 9% more relative airtime than their share of participants would predict. In the first — the un-recorded, un-watched default — the well-documented pattern is the opposite: women speak about 25% less than men in a typical meeting.
That's not a small effect, and it's not really a story about notetaking software. It's a story about what happens to human behavior the moment people believe they're being watched — and it turns out one of the most reliable ways to trigger that belief is to put a small, visible AI in the room.
The study, and the gap it reverses
The finding comes from Read AI's analysis of meeting data, covered by Forbes contributor Michelle Travis in a February 2026 piece on gender and meeting airtime. Read AI's own notetaker sits in on a huge number of real corporate meetings, which gave the company an unusually large, naturalistic dataset to look at a specific question: does the presence of an AI notetaker change who talks?
The baseline they were comparing against is not new or controversial. The finding that women speak substantially less than their numeric share of a meeting would predict — commonly cited around a 25% gap — has been replicated across academic and workplace research for well over a decade, going back to foundational work like Karpowitz and Mendelberg's research on gender and group deliberation. Airtime in meetings isn't neutral; it functions as a visible signal of authority, competence, and "who's really running this." People who interrupt more, speak first, and hold the floor longer are consistently read as more confident and more central to decisions — regardless of whether what they said was actually the most useful contribution in the room.
So the baseline is: without intervention, meetings default to a pattern where men get more of the signal that reads as leadership. The Read AI finding is that this default isn't fixed — it's sensitive to whether people believe they're being observed. When a notetaker was present, the gap didn't just shrink. It reversed, with women getting more relative airtime than a simple headcount would predict.
Why a bot in the corner changes what a room does
The mechanism the study points to is a version of something psychologists have studied for decades under different names — self-monitoring, social facilitation, the observer effect. The short version: people behave differently when they know their behavior is being recorded and will be reviewed, even if no one is watching live. It's the same instinct behind why a classroom gets quieter when a video camera is visibly recording, or why people drive more carefully past a sign that says a speed camera is ahead, whether or not the camera is actually loaded with film.
In a meeting, an AI notetaker makes that observer effect concrete and personal. It's not an abstract policy that recordings might happen somewhere. It's a name or a bot icon sitting in the participant list for the whole hour, and everyone knows a summary is coming afterward that will attribute talk time, decisions, and action items to specific people. That changes the incentives in real time. The person who might otherwise talk over a colleague or run long on a tangent gets a small, persistent nudge to notice it — because the record will notice it too. We've written before about this same mechanism from the other side, in how knowing you're being recorded changes what you say in a meeting — the gender-airtime finding is really a specific, measurable instance of that broader effect.
The gap in who gets heard isn't fixed biology or fixed culture — it's a default that assumes no one is watching. Introduce a credible observer, and the default moves.
This lines up with something researchers on airtime and status have argued for a while: the gender gap in meeting speech isn't primarily about women having less to say. It's about interruption patterns, about who feels licensed to speak first and speak longer, and about the fact that dominant talkers rarely self-correct without a reason to. An AI notetaker gives everyone in the room — including the person who's used to holding the floor — a concrete reason to self-correct, without anyone having to say a word about it out loud.
The part of the mechanism that matters more than the headline number
Here's the detail worth sitting with: almost every notetaker used in studies like this one is a cloud bot. Read AI, like Otter, Fireflies, and most of the category, works by dialing into the video call as a literal participant — a named tile with a face or logo, visibly present in the grid for the whole meeting, uploading audio to a server to be transcribed and summarized. That visibility is very likely doing most of the work here. You can't get an observer effect from an observer nobody notices. The bot has to be seen to change behavior.
But a visible cloud bot buys that behavioral effect at a real cost. Every participant's voice is now leaving the room and going to a third-party server. Someone has to be added to a meeting as an actual attendee before it starts. There's an account behind it, a data-retention policy to trust, and — as some notetaker vendors have found out the hard way — real legal exposure around recording without proper consent or storing biometric voiceprints without disclosure, which is exactly the kind of dispute we cover in what the AI notetaker voiceprint lawsuits are actually about. Getting a room to behave better is a genuinely good outcome. Getting it by routing everyone's voice through a cloud server and a bot with a seat at the table is a tradeoff, not a free lunch — and it's one reason so many teams still hesitate before turning a notetaker on, a hesitation we go into in the etiquette of asking a room to be recorded.
Same self-monitoring effect, without the bot or the upload
We want to be precise about what we can and can't claim here. Meetly was not part of the Read AI study, and we're not going to pretend it was. What we can say is that the mechanism the researchers describe — people self-monitoring because they know their speech is being captured and will be reviewed — doesn't actually require a cloud bot. It requires that people believe, correctly, that the meeting is being recorded and that a record will exist afterward. A visible cloud participant is one way to create that belief. It is not the only way.
Meetly runs the recording, transcription, and summarization entirely on the iPhone of the person who opened the app — no bot joins the call, and no audio leaves the device. Nobody has to accept a calendar invite for a robot. There's no cloud upload, no third-party account holding the transcript, no separate consent flow for a company none of the other participants chose. And when the person running Meetly says at the top of the call, "I'm recording this so I don't have to type while we talk" — which is good practice regardless of tool — the room gets the same information a Read AI-style bot tile would have given it: this is being captured, and a summary is coming. That's the input the observer effect actually runs on. The delivery mechanism — cloud bot with a visible seat, or a phone on the table with the owner's word for it — is a separate question from whether the behavioral effect fires.
This is an inference from the mechanism the study describes, not a claim that we ran the same experiment. It would take a dedicated study to prove that an on-device recorder produces the identical 9% shift Read AI measured. But the logic doesn't require a bot with a face in the meeting grid — it requires people knowing, credibly, that their words are being kept and will be looked at again. A private, on-device notetaker can deliver that belief just as honestly as a cloud one can, without asking every participant to trust a third party's servers with their voice to get it.
Get the self-monitoring effect without the cloud bot
Meetly records and transcribes meetings entirely on your iPhone — no bot joins the call, no audio leaves the device, no third-party account. Say it out loud at the start of the meeting, and the room gets the same signal that changes how people talk — without anyone's voice going anywhere but your own phone.
Download MeetlyWhat this means beyond one Forbes-covered study
The bigger point isn't "add a notetaker and your gender gap fixes itself." It's that a lot of meeting dynamics people treat as fixed personality traits — who dominates, who defers, who gets interrupted — are actually defaults that hold specifically because nobody's watching closely enough, in the moment, to notice or care. Related work on why certain recurring meetings persist long after they've stopped earning their slot, which we covered in does this meeting need to exist, points at a similar root cause: without a feedback loop, meetings drift toward whatever pattern is easiest for the people already comfortable with the current one. Airtime is no different. Left alone, it drifts toward the people already used to taking it.
- The baseline gap is well-established — women speaking meaningfully less than their share of a meeting predicts is a long-replicated finding, not a one-off statistic.
- The reversal is new and specific — Read AI's 2026 data (via Forbes/Michelle Travis) found roughly 9% more relative airtime for women specifically when an AI notetaker was present.
- The likely mechanism is self-monitoring — believing you're being recorded and reviewed changes behavior, a well-studied observer effect, not something unique to AI.
- Visibility is probably doing the work — the notetakers studied are cloud bots with a visible seat in the call, which is what makes the recording impossible to forget about.
- The privacy cost is real — visible cloud bots mean uploaded audio, third-party accounts, and consent questions for every participant, not just the organizer.
None of this makes the underlying imbalance disappear when the recording stops. The Read AI finding is about behavior in the presence of an observer, not a durable shift in who gets deferred to once the notetaker is off. But a mechanism that reliably rebalances a meeting for its duration is still worth taking seriously — especially if you can get it without shipping every participant's voice to a server to do it.
Record the meeting. Keep the voice on the phone.
Meetly gives you a full on-device transcript and summary — private by architecture, no bot in the call, nothing uploaded. Download it free and see what changes in the room when everyone knows the meeting is being kept, honestly, without a stranger's server involved.
Download MeetlyAI notetakers and the meeting gender gap: frequently asked questions
Do AI notetakers reduce the gender gap in meeting speaking time?
A 2026 study by Read AI, covered by Forbes, found that when an AI notetaker was present in a meeting, women got roughly 9% more relative airtime than their share of participants would predict — reversing the well-documented baseline where women typically speak about 25% less than men in meetings without one. The likely mechanism is an observer effect: people self-monitor more when they know their speech is being captured and reviewed.
Why do women talk less in meetings on average?
Research going back over a decade, including work on gender and group deliberation by researchers like Karpowitz and Mendelberg, has repeatedly found that women speak substantially less than their numeric share of a meeting predicts. Airtime functions as a status and leadership signal, and interruption patterns and floor-holding tend to favor whoever is already used to taking the floor — a dynamic that isn't self-correcting without some form of feedback or observation.
What is the observer effect in AI meeting notetakers?
It's the behavioral shift that happens when people know a meeting is being recorded and will be reviewed afterward — similar to how people slow down near a visible speed camera. In meetings, a visible AI notetaker (a bot tile in the call, a known summary coming afterward) makes that awareness concrete, which appears to push people toward more balanced, self-monitored speaking behavior, including noticing when they're dominating airtime or interrupting.
Does Meetly change meeting behavior the same way as cloud AI notetakers?
Meetly wasn't part of the Read AI study, so we can't claim it produces the identical measured effect. What we can say is that the mechanism researchers describe — self-monitoring because you know you're being recorded — depends on people believing a record exists and will be reviewed, not specifically on a cloud bot joining the call. Meetly delivers that same belief through an on-device iPhone recording with no bot and no cloud upload, which is a different privacy tradeoff for the same underlying behavioral trigger.
Do you need a visible bot in the call for an AI notetaker to change behavior?
The studies to date have mostly used cloud notetakers that join as a visible participant, so that's the evidence we have. Visibility plausibly matters because an observer effect needs people to notice the observer. But announcing at the start of a meeting that it's being recorded — whether by a cloud bot or an on-device app — communicates the same information that a bot tile does, which is the input the effect appears to run on.