Do AI Notetakers Make Meetings Better, or Just Easier to Skip?

By Victor Hoang · Updated Oct 13, 2026

Most leaders who roll out an AI notetaker expect two things: fewer people scribbling notes, and fewer people needing to attend. The second expectation is the interesting one. If a summary lands in your inbox ten minutes after the call, why sit through the call?

For some meetings, that's exactly the right question, and the honest answer is that you shouldn't. For others, it's the start of a slow decay. People skip, the summary becomes the record, and the decisions that used to be made by a group of people looking at each other become something that was "shared in a recap."

Our view is simple. AI notes transfer information well, but they don't transfer commitment. And recording a conversation changes what people are willing to say in it. So the useful question isn't whether to use notetakers. It's which meetings exist to move information, and which exist to create ownership. Use the tool to kill the first kind. Protect the second kind, live and unrecorded.

Why the Notetaker Question Is Really a Meeting-Purpose Question

Meetings were already in trouble before AI showed up. Researchers writing in Harvard Business Review surveyed 182 executives and, as Boston University summarizes it, found that a majority of senior managers believe too many meetings are unproductive, keep them from getting work done, and don't bring teams closer together. Nobody needed a new technology to feel that.

What the notetaker changes is the cost of the information half of a meeting. Before, a status update required a live audience because the only way to get the news was to be in the room. Now the news can be captured, summarized and searched. That's a real gain, and it's why we think leaders should be aggressive about it. A weekly meeting where eight people take turns saying what they did is a document wearing a calendar invite.

But a meeting has a second job that has nothing to do with information. People hear a decision made, watch who objects, see who stays quiet, and leave having publicly agreed to something. That's commitment, and it's produced by presence and exposure, not by text.

Key Facts

Key Facts

• Microsoft's 2025 Work Trend Index telemetry found that meetings after 8 pm are up 16% year over year, and that half of all meetings fall into two windows, 9 to 11 am and 1 to 3 pm.
• Korn Ferry reported in August 2026 that a loss of candor is the biggest concern among leaders who are questioning whether recording every meeting is stifling creativity.
• Federal law requires the consent of at least one party to record a conversation, but about 11 states primarily require the consent of all parties.

The Two Jobs a Meeting Does

Here's the mental model we use, and it's our own framing, not an established standard. Every meeting does one or both of two jobs.

Job one is transfer. Information moves from people who have it to people who need it. Status updates, announcements, read-outs, training walkthroughs and most all-hands sessions are mostly this.

Job two is commitment. People who will have to act together decide something, test each other's reasoning, and take on ownership in front of one another. Hiring decisions, budget trade-offs, pricing calls, incident reviews and strategy disagreements are mostly this.

Call it the Transfer-or-Commit test: before you let anyone record, summarize or skip a meeting, ask which job it exists to do. If the honest answer is transfer, an AI summary does it better and faster than a calendar block. If the answer is commit, a summary is a receipt for something that may not have happened.

The mistake we see is applying one policy to both. Companies that mandate a notetaker on every call get the efficiency on status meetings and the damage on decision meetings. Companies that ban notetakers everywhere keep the candor and pay the transfer cost forever. Neither is a strategy.

Why Summaries Don't Carry Commitment

It's tempting to think the problem is accuracy: get a better summary and absence stops mattering. We don't think so, for three reasons.

First, a summary compresses away the part that creates ownership. It tells you that the pricing change was approved. It doesn't tell you that the head of sales hesitated, that finance asked a question nobody answered, or that the decision was made over a quiet objection. Those details are exactly what the people carrying it out need, and they live in the room.

Second, reading isn't agreeing. Someone who skims a recap hasn't been asked to say yes. When they later deliver something that doesn't match, they can truthfully say they never signed up for it. We've made a related point about decisions that should never be delegated to AI: the value of a decision partly lies in a named human choosing it.

Third, effortless capture can lower attention. A 2025 experiment accepted at the ACM's CSCW conference had 30 participants take notes on lecture videos under three levels of AI support. The group using the most automated, structured notes scored lowest on the post-test, while moderate AI support scored highest, even though people preferred the automated option because it took less effort. That was a lecture setting, not a leadership meeting, so treat it as a signal rather than proof. But the mechanism is plausible for any meeting: when the tool does the remembering, the person does less thinking.

There's also a verification cost nobody budgets for. If a summary has to be read, checked and corrected before anyone acts on it, some of the saved time comes back out. We covered that pattern in the verification tax on AI time savings.

What Recording Does to Candor

The second half of the thesis is about what people say once the red light is on.

The research base here is thinner than the confident commentary suggests. We haven't found a controlled study that directly measures speaking time or candor in recorded versus unrecorded business meetings, and we'd be wary of anyone who quotes a precise percentage. What we do have is a well-established foundation and a strong signal from practice.

The foundation is Amy Edmondson's work on psychological safety. Her 1999 paper in Administrative Science Quarterly, based on a study of 51 teams at a manufacturing company, defined it as a shared belief that the team is safe for interpersonal risk taking and linked it to team learning behavior. Candid disagreement, admitting a mistake and floating a half-formed idea are all interpersonal risks. A permanent, searchable, shareable transcript raises the price of each of them. Nobody has to be punished for the effect to occur. They only have to imagine the clip being replayed in a performance review or forwarded to a customer.

The signal from practice is that leaders feel it. Korn Ferry's August 2026 piece describes loss of candor as leaders' biggest worry about recording, specifically the loss of spontaneous thought and the willingness to challenge ideas. That's a statement about what leaders fear, not a measurement of what happens, and we'd read it that way. But it matches what most experienced managers will tell you privately: the best version of a hard conversation is rarely the one you'd want on file.

If your team already goes quiet, a recorder won't help. We've written about why teams stay silent in meetings and how to design a better meeting culture, and both are worth reading before you decide where recording belongs.

The Legal Layer You Can't Skip

Consent is not a formality, and it's the part leaders most often assume is handled by the vendor.

Under federal law, the baseline is one-party consent. The Reporters Committee for Freedom of the Press explains that federal law requires the consent of at least one party to record in-person, telephone or electronic conversations, that it sets the minimum across the country, and that states may impose stricter rules. About 11 states primarily require all parties to consent. For calls that cross state lines, the same guide advises assuming the stricter law will apply.

That's the floor. The practical risk sits in how notetaker software behaves. In August 2025, a consumer named Brewer filed a proposed class action against Otter.ai in the Northern District of California. As Fisher Phillips summarized, the complaint alleges that the app records meetings without the consent of all participants, including people who aren't account holders, and that it keeps conversation data to train its speech recognition. Those are allegations, no court had made findings when the firm wrote, and the company had not yet responded. We cite the case as a warning about the category, not as a verdict on one vendor.

For an executive, the takeaway is operational. A bot that joins a call because the host connected a calendar is not the same as a participant who tells everyone they're recording. Your policy should cover disclosure at the start of the call, external guests, where recordings are stored, who can search them, and whether the vendor trains on your data. Fisher Phillips' guidance runs along the same lines and also advises legal, HR (human resources) or compliance review before recording privileged or sensitive HR discussions. This is general information, not legal advice, and the rules in your states and in your guests' states are what apply. It sits naturally inside a wider AI governance program.

A Decision Table: Record, Summarize, or Keep It Live

Here's how we'd apply the test. Treat it as a starting point and adjust for your industry and your legal advice.

Notice what the table doesn't say. It doesn't say commit meetings should have no written trace. They need a trace, but a deliberate one: a short decision record written by a human who owns it, naming what was decided, who owns it, and what dissent was raised. That's not a transcript. It's closer to a signature.

Using Notetakers to Kill Status Meetings

The upside is large, and most companies leave it on the table because they add the notetaker without removing the meeting.

The sequence that works is subtractive. Pick the recurring meetings that score as pure transfer. Replace the live version with a written update, or with a recorded session that people can watch at their own pace. Use the AI summary as the distribution mechanism. Then take the calendar time back and spend it where commitment is built.

This matters more than it looks. When Microsoft's telemetry shows meetings creeping into the evening and half of all meetings clustered into two daily windows, the issue isn't just volume. It's that every transfer meeting crowds out the scarce live time that decision forums need. The coordination cost of getting the right people into one room is real, and we've looked at why coordination is the hidden cost in operations. Every status meeting you retire makes the next decision forum easier to schedule.

It's also worth pairing this with a clear-eyed look at which of your meetings were never worth having. Our piece on cutting pointless meetings is a useful place to start the audit, and a practical guide to AI meeting notes and summaries covers the tooling side.

The Risk of Making Meetings Easier to Skip

Skipping is the failure mode the title warns about, and it's not hypothetical. When people know a summary exists, attendance in the meetings that matter starts to feel optional too. A few patterns to watch for:

• The empty decision room. Six invited, three present, and the rest "will catch the recap." Whatever the three decide, the other three haven't agreed to.
• The recap as record of truth. Disputes end with "the summary says," and nobody remembers whether the summary captured the nuance.
• The performative meeting. When everyone knows it's recorded, people speak for the transcript, and the real discussion moves to side channels, which are less visible and harder to govern.

The answer isn't to ban the tool. It's to be explicit about attendance. For commit meetings, attendance is required and the recording is off. For transfer meetings, attendance is optional and the recording is on. Saying that out loud removes the ambiguity that lets the first category erode into the second.

What to Do Next

For a COO, chief of staff or head of operations, this can be done in a month.

1. List your recurring meetings. Take the top 20 by total attendee hours. For each, write one word: transfer or commit.
2. Retire or convert the transfer meetings. Replace live status calls with a written update or a recorded session. Aim to remove at least a third of them in the first quarter.
3. Name your protected forums. Decide which decision, review and coaching meetings are live and unrecorded, and say so in writing.
4. Write the consent rule. Disclose at the start of every recorded call, handle external guests explicitly, and have counsel review rules for your states and your guests' states.
5. Set vendor terms. Ask where recordings are stored, who can search them, how long they're kept, and whether your data trains the vendor's models. Fold the answers into your AI governance basics.
6. Require a human decision record. After every commit meeting, one named owner writes three lines: decision, owner, dissent. Don't use an AI-generated recap as the record of a decision.
7. Check after 60 days. Ask the people in your protected forums whether they're speaking more freely, and ask the people you freed from status meetings what they did with the time. If the second answer is "nothing in particular," you've saved hours but not value.

So do AI notetakers make meetings better? They make the information half of meetings better, and the quality of that half is no longer a reason to gather people. Whether they make the commitment half worse depends on whether you decide, in advance, which meetings are which.