The Meeting Notes Trap: When Automated Summaries Change What People Say

You turned on automatic meeting notes and the meetings got worse. Not obviously worse. People still show up, still talk, still agree things. But the half formed thought that used to lead somewhere has stopped appearing.

This is a real and predictable effect. When people know a meeting is being captured and summarised, they speak in finished sentences, avoid saying they do not know, and stop testing ideas out loud. You gain a record and lose the thinking that used to produce the decisions worth recording.

What actually changes

Three specific behaviours shift, and they are worth naming because they are easy to mistake for something else.

People stop speculating. "What if we just stopped doing this entirely" is a useful sentence that nobody wants attributed to them in a summary sent to twelve people.

People stop admitting gaps. Saying "I do not know, I have not looked at that" is how a team finds out where its blind spots are. In a captured meeting it reads as unpreparedness.

People start performing for the absent reader. The meeting quietly acquires an audience that is not in the room, and everyone begins addressing it. Discussions get longer and less honest at the same time.

Why "just get used to it" does not work

The common response is that people will adjust once recording feels normal. Some do. But the ones who adjust fastest tend to be the ones most comfortable with visibility, and the ones who stay careful are often the people closest to the problems you need surfaced. It runs on the same nerve as telling staff an agent is doing part of their job: what you say about the tool decides how people behave around it.

Habituation also runs the wrong way over time. As summaries accumulate and get quoted back in later contexts, the sense that anything said might resurface gets stronger rather than weaker. They are also the first thing read by a new hire joining alongside an agent that already knows the job. People do not relax. They get better at saying nothing.

Separate the thinking from the deciding

The most reliable fix is structural. Split meetings into ones that produce a record and ones that do not, and be explicit about which is which at the start.

Exploratory conversations, where the point is to work out what the question even is, run without capture. Decision meetings, where the point is to settle something, run with it.

This works because it gives people a place to be uncertain out loud. It fails when the split is nominal, so it needs to be visible: the capture is off, the screen says it is off, and nobody quietly writes it up afterwards anyway.

Capture decisions, not transcripts

The second fix is to narrow what the summary contains.

A full narrative summary of who said what is the thing that changes behaviour most. A short record of what was decided, who owns it, and what the deadline is changes behaviour hardly at all, because nobody is nervous about being recorded as having agreed to do a task.

Ask for four things from an automated summary: decisions, owners, dates, and open questions. Leave out attributed opinions, leave out the debate, leave out anything that reads like a character assessment. Writing that instruction down once is exactly the sort of thing that belongs in a prompt library or a written procedure.

Open questions are worth keeping because they are the one part of the discussion that has value later and carries little social risk.

Control who can read it, and say so

Half the caution in a captured meeting comes from uncertainty about the audience. People behave differently when they know the note goes to the six attendees than when they suspect it might end up in a board pack.

Make the audience explicit and enforce it. This is easier when meeting records live on the same set of records as projects and people rather than in a separate tool with its own sharing rules. In SicherOne, agent produced output goes to a human before it ships, which also means someone can catch a summary that has drifted from decisions into commentary before anyone else reads it.

State the retention period too. "Deleted after ninety days" changes how a room feels more than any reassurance about intent.

Let people correct the record

Give attendees a short window to amend a summary before it is final, and treat corrections as normal rather than as a complaint.

This matters more with automated notes than with a human note taker, because a model can misattribute a position to the wrong person in a way that reads entirely plausibly. The person who was misquoted is often the last to see it.

A visible correction process also reduces defensiveness in the room, because being misrepresented stops being permanent.

The test to run

Sit in a meeting with capture on and ask one question: has anyone said "I do not know" or "this might be a bad idea" in the last twenty minutes?

If not, the summary you are collecting is a record of a conversation people had for the record. It will be accurate, tidy, and missing the part you actually needed.