Should you let AI notetakers record your client meetings?

Picture the most sensitive meeting your firm will hold this week. Now picture an AI assistant quietly sitting in it, recording every word and emailing a summary to whoever it decides should get one. For a lot of firms that is not a hypothetical, it is the default, and nobody chose it. An AI notetaker in a client call, a candidate interview or a deal conversation is a decision the firm should make on purpose, not a default to leave running. Three things sit inside it: who agreed to the recording, where the recording ends up, and who is still accountable for whether the written record is right. This is general guidance, not legal advice, so use the law below to check your own position rather than as a ruling.

The guest you did not introduce

The tools are familiar. Microsoft 365 Copilot, Otter, Fireflies, Granola, Zoom's AI Companion and others now join calls on their own, transcribe in real time and produce notes without anyone lifting a finger. The convenience is real. So is the failure mode. In 2024 an AI researcher finished a video call with a group of investors, then opened his inbox to find an Otter transcript that carried on past the point he had left, capturing the investors' private discussion of what he described as strategic problems and questionable numbers. Those extra paragraphs, by his account, killed the deal, as NPR reported. Otter is now the subject of a US class action alleging it recorded private conversations "deceptively and surreptitiously", and that by default its notetaker does not ask everyone in a meeting for permission and can join automatically when the host has linked it to Zoom, Google Meet or Teams. The company says some 25 million people use it, across more than a billion meetings. For a leader, the word worth dwelling on is "default". These tools record because they are set to, not because anyone in the firm decided they should.

Why this is a leadership question, not an IT setting

Strip away the brand names and the same shape sits under each tool. A participant you did not introduce is now in your most confidential conversations, behaving the way it was configured to behave rather than the way you would have chosen. Call it the silent guest. It presses on three things a professional-services firm actually sells: confidentiality, trust, and an accurate account of what was said. Take them in turn.

Where does the recording go?

A transcript is more than a private note. It is a copy of a client's matter, a candidate's interview or a deal's terms, made automatically and sent somewhere. The questions to ask of any notetaker are dull and decisive: what does it capture, where is the recording stored, who inside and outside your firm can reach it, and is any of it used to train the vendor's model. Many tools keep transcripts in the cloud, and some use meeting content to improve their systems unless you turn that off. For a firm that has promised a client confidentiality, the honest answer to "who holds a copy of this conversation" should never be "a software company we never vetted".

Did the other side agree to be recorded?

The client or candidate across from you did not necessarily agree to an AI recording them, and may not even know it is happening. That matters as courtesy and trust before it matters as anything else, because people speak differently when they find out afterwards that a machine kept the tape. There is a legal dimension too, and it deserves a plain, careful word. In the UK, recording a meeting captures personal data, so under UK GDPR and the Data Protection Act 2018 people should generally know they are being recorded and why, and your existing duty of confidentiality still governs where that recording sits. The ICO publishes guidance on this, which is its expectation and interpretation rather than a statute, and there is no separate UK AI law that changes the basics. The recording-consent lawsuits you may have read about are American, decided under American rules, so do not treat a US case as your obligation. The point that travels is simpler. A candidate who learns later that an AI logged their interview has found out something about how your firm treats people, and it is not flattering. Tell them first.

Who owns the record?

An AI summary reads with calm authority, which is exactly the problem. These tools mishear names, attribute a remark to the wrong person, flatten a hedged "we might" into a firm "we will", and now and then log an action nobody agreed to. Then the tidy note circulates as the record, and a month later it is what people remember deciding. Ethan Mollick's argument about working with AI lands cleanly here: the skill that matters is oversight, and it is your own expertise that lets you catch an output that is wrong and correct it (Management as AI superpower). A machine can draft the minutes. A named person still has to read them against what was actually said and sign them off before they travel. That is the same verification habit that stops AI mistakes reaching clients in written work, applied to the record of a meeting.

The rule, in four lines

You do not need a committee for this. You need one short position your people can follow without thinking, written into your wider AI policy rather than left to each person's settings. Four decisions cover it.

First, decide whether a notetaker is on by default, and for which meetings. For sensitive client, candidate or deal conversations, off by default is the safer starting point.

Second, agree before you record. Whoever hosts asks the room in plain words at the start, and a no means off.

Third, decide where transcripts live and for how long, and check whether the tool uses your meetings to train its model. If you cannot answer that, you have not chosen the tool, it has chosen you.

Fourth, name who checks the minutes. One person reads the AI summary against the meeting and signs it off before it circulates.

Getting to a rule like that is less about the tools than about a leader understanding what AI is actually doing in their own meetings, which is the work AI Lessons for Leaders is built around: not a tool demo, but the judgement to decide where AI belongs in a firm's work and where it does not.

Here is the test. If a good client asked you today what happens when an AI takes notes in their meeting, could you answer in one sentence, and would it reassure them? If yes, you are ahead of most firms. If you had to stop and wonder who holds the recording, you have found the hour of work worth doing before the next meeting starts. When you want a hand getting there, book a discovery call.

FAQ

Should you let AI notetakers record your client meetings?

Only on a rule you have set, not on the tool's default. For sensitive client, candidate or deal conversations, start from off by default, ask everyone present before recording, know where the transcript is stored and whether it trains the vendor's model, and have a named person check the summary before it circulates. The convenience is real, but it should sit inside a position your firm has chosen.

Do you have to tell people an AI is recording the meeting?

As a matter of trust, yes, tell them first, because people speak differently when they discover a recording afterwards. In the UK there is also a data-protection dimension, since recording captures personal data and people should generally know they are being recorded and why, and your confidentiality duties still apply to where the recording sits. This is general guidance rather than legal advice, and the much-publicised consent lawsuits are American and decided under American law, so take proper advice on your own situation rather than treating a US case as your rule.

Can you rely on the AI meeting summary as the official record?

Not without a human check. AI notetakers mishear names, misattribute comments and can turn a maybe into a firm commitment, and the tidy summary then circulates as what was decided. Treat the AI version as a first draft, have one named person read it against what was actually said, and only then let it stand as the record.