Field note
Using AI meeting notetakers without breaking client confidentiality
Before you let Otter, Fireflies, or Zoom AI Companion sit in on a client call, know what happens to the recording after it ends.
AI meeting notetakers like Otter, Fireflies, and Zoom AI Companion can join client calls safely, but confidentiality is not a setting you accept by default. Before you turn one on, you need to disclose it to everyone on the call, check what the vendor keeps and trains on, and keep the tool out of privileged or NDA-bound meetings entirely.
Say it out loud
Name the tool, every call, not just consent states
Check the defaults
Retention window and the model-training toggle
Turn off what you can
Shorten retention, disable training
Skip privileged calls
Legal, board, anything under NDA
What actually happens to the recording after the call ends
Most AI notetakers are cloud services, not local apps. The audio, video, and transcript usually leave your laptop and sit on the vendor's servers, sometimes for months, sometimes being used to improve the vendor's own AI models unless you turn that off. Zoom is one of the few that states plainly it does not use customer audio, video, chat, or transcripts to train its own or third-party models, and shows meeting participants an on-screen indicator when AI Companion is active. Other tools default to broader data use, so the safe assumption with any new notetaker is that it keeps and trains on your data until you find the setting that says otherwise.
Do you need everyone's consent before you hit record
Federal law only requires one person on the call to know it is being recorded. California, Connecticut, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania, and Washington require every participant's consent instead, and the law treats an AI notetaker exactly like a human holding a recorder: the technology does not change the obligation. If any client, prospect, or partner on the call is in one of those states, you need an explicit yes before the bot joins, not a quiet assumption that staying on the call counts as agreement.
What to check before you let a notetaker into a client call
Treat the notetaker like any other vendor you are handing client data to, because that is what it is. Say out loud, every call, that an AI tool is recording and taking notes, and name the tool. Open the vendor's settings and find the retention window and the model-training toggle, then shorten the first and switch off the second wherever the option exists. Check where the data is actually stored, since some tools route it through servers outside the country your client expects their information to stay in. None of this takes more than a few minutes once you have done it once.
Which meetings should never go through a third-party notetaker
Some conversations are not yours to hand to a vendor, whatever the settings say. Calls covered by attorney-client privilege, board discussions, anything already under an NDA that restricts third-party processors, and conversations about a client's own confidential customer data belong off any tool you do not fully control. For those, take notes yourself, or use a notetaker that stays inside infrastructure your organisation already runs and trusts, and skip the convenience the rest of the time: it is not worth the exposure.
An AI notetaker has already sat in on a meeting for roughly one in three employed Americans, and most were not reliably asked first.
Sources
The takeaway
Before your next client call: say out loud that an AI notetaker is recording, open its settings and turn off model training and shorten retention, and keep it out of any meeting covered by privilege or an NDA.