I keep seeing the same awkward Microsoft Teams moment. A meeting is running, transcription is on, everyone knows transcription is on because Teams literally told everyone, and yet somewhere around minute 37 the conversation becomes a little more interesting. Someone mentions an employee, someone shares a personal situation, someone gives a very honest opinion about a colleague or says something confidential about a customer. And then comes my favorite sentence: “Can we keep this between us?” Well… Teams would also like to be part of that secret. 😂 Because the transcript is still running. Five minutes later the organizer gets a message: “Could you please delete the transcript?” I see versions of this happening far too often and the interesting thing is that this isn’t really a Teams problem. It is a human behavior, AI literacy and information governance problem.
For years we treated meetings as temporary spaces. You said something, the people in the room heard it, somebody wrote down three completely random notes, and life continued. Recording already changed some of that, but transcription and AI take it much further. The moment transcription is running, spoken conversation can become persistent, searchable and attributable information. That transcript can subsequently support recap and Copilot experiences depending on licensing, policies and meeting configuration. So perhaps we need to stop thinking of transcription as a handy little meeting feature. You are creating data. And once you start looking at it that way, some of our meeting behavior suddenly looks quite strange. Imagine typing into Teams: “I probably shouldn’t write this down, but let me tell you something personal about Sarah…” You probably wouldn’t press Enter. Yet people say the equivalent in transcribed meetings every single day. Transcription doesn’t make a bad conversation safe. It just makes it searchable.
There is also quite a bit of confusion about who actually “owns” the transcript. I have heard people assume it belongs to whoever joined first, whoever clicked Start transcription, whoever is presenting or whoever happens to be facilitating. For a normal scheduled Teams meeting, that isn’t the right mental model. Microsoft stores the transcript in the meeting organizer’s OneDrive for Business. If the meeting is recorded and transcribed, the recording is normally stored there too. Channel meetings work differently and use the team’s SharePoint environment. So being first in the meeting doesn’t magically make you the transcript owner, and pressing the transcription button doesn’t either. The organizer role and meeting configuration matter. But there is an even more important distinction: being allowed to start transcription and being allowed to access the resulting information are not the same thing. And that gives us a much better question than “Who pressed the button?” Who should have access to the digital memory of this meeting? That is where this becomes governance.
I love transcription. I love meeting recap. I love Copilot. I absolutely do not want to return to the days when my meeting notes consisted of three words, two arrows and a star next to something I apparently thought was very important. 😂 But that doesn’t mean every meeting needs a permanent memory. A project meeting? Great. A weekly team meeting? Probably useful. A brainstorming session? Sure. But a conversation about someone’s performance, health or personal circumstances? An HR case? A sensitive customer situation? Legal matters? Security incidents? Executive discussions? Think first, transcribe second. The question should never just be “Can Teams transcribe this?” Of course it can. The much better question is: “Should this conversation become persistent organizational data?”
And that also means the job of the meeting organizer has changed. Being an organizer used to mean finding an impossible gap in twelve calendars, sending an invite and spending the first seven minutes asking whether everyone could see your screen. Now, if you organize meetings with recording, transcription or Copilot, part of that responsibility is information governance. Who can start transcription? Who can access it afterwards? Are external participants involved? Could sensitive information be discussed? Do we even need transcription? Microsoft gives organizations controls for many of these things, but Microsoft cannot decide whether the conversation you are about to have should be transcribed. That bit is still ours.
And that brings me to perhaps my biggest frustration with this whole discussion: please take responsibility for your own mouth. 😂 Yes, organizers have responsibilities. IT has responsibilities. Security and compliance have responsibilities. Organizations need policies. But if Teams tells you that transcription is running, you also have a responsibility for what comes out of your mouth. We cannot create a culture where people say inappropriate or unnecessarily sensitive things in a transcribed meeting and then make it someone else’s job to clean up the digital evidence afterwards. Especially when you are talking about another person. Personal circumstances, medical information, performance concerns, rumours, salary information, opinions about colleagues or confidential customer information don’t suddenly become appropriate because you spoke them instead of typing them.
There is actually a wonderfully simple test for this: Would I type this exact sentence into Teams and press Enter? If the answer is “ABSOLUTELY NOT 😳”, perhaps don’t dictate it into a transcript either. And we should become much more comfortable challenging each other here. “Hey, transcription is still running. Should we stop it before we continue?” should become a completely normal sentence in modern meetings. That isn’t being difficult. That is digital professionalism.
There is another layer to this that becomes even more important as AI enters our everyday work. Historically, meeting governance was mostly about attendance. Who is invited? Who joined? Who is external? But now we have to think about something else too. Someone might legitimately need to join ten minutes of a sixty-minute meeting. That doesn’t automatically mean they need persistent access to everything discussed during the other fifty minutes. So the question is shifting from “Who can enter the room?” to “Who should have access to the memory of the room?” Those are not the same thing. Once AI enters the equation, I think we need to distinguish three concepts much more clearly: meeting access ≠ information access ≠ AI access. That is going to become an increasingly important part of how organizations design their AI governance.
And yes, transcripts can be deleted. Sometimes that will absolutely be the right thing to do. But “please delete the transcript” cannot become our governance strategy. If your standard operating model is “we transcribe everything and delete it when somebody says something they shouldn’t have said,” you don’t have a deletion problem. You have a behavior and governance problem. The better model starts before the meeting. Decide whether transcription is appropriate, configure access and make people aware of what is happening. During the meeting, stop transcription before deliberately moving into sensitive territory where appropriate. Afterwards, understand where the information lives and who can access it. And throughout all of that, perhaps the most sophisticated security control available to us is still the one humans have had for thousands of years: think before you speak.
IT has some homework here too, and I don’t think we solve this with another mandatory e-learning where everyone clicks “Next” seventeen times while answering emails on another screen. Organizations need an actual meeting governance model. Standard collaboration meetings might allow transcription. Confidential internal meetings might require more restricted access. Highly sensitive HR, legal, security or executive conversations might have transcription disabled by default unless there is a clear business reason and appropriate controls are applied. The exact model will differ between organizations, but “everyone just figures it out themselves” is not an enterprise AI strategy.
And this is where, for me, the story becomes much bigger than Teams transcription. A transcript used to sound like a boring text document sitting somewhere after a meeting. That mental model is already outdated. Transcripts can support intelligent recap, summaries and Copilot experiences. Our meetings are becoming machine-readable organizational memory. We used to ask, “Who heard what I said?” Now we also need to ask, “What data did my words create, who can access it and what can AI do with it?” That is AI literacy too. It isn’t only knowing how to write a great prompt or getting Copilot to summarize a document. It is knowing when information should not become AI-accessible data in the first place.
So perhaps we need one very simple new meeting rule: if you wouldn’t want a sentence attributed to your name, stored as organizational information and potentially processed by the AI experiences permitted in that meeting, don’t say it while transcription is running. And if the conversation genuinely needs to become sensitive? Stop the transcript and continue the conversation. We are giving our meetings a memory. Maybe it is time we started behaving like they have one.
