Years ago I read a book that argued against taking notes in meetings. The claim was simple. Listen hard enough and your brain will hold onto what actually matters. Write everything down and you outsource the filtering job to a notebook you will probably never reopen. I took that seriously enough to change how I sit in meetings. I do not take notes. I watch the person talking. I catch the pause before an honest answer, the tone that says someone cares more than their words let on, the moment enthusiasm becomes real instead of performed. None of that survives in a transcript. A transcript hears what was said. It has no idea what any of it meant to the person saying it.

The proliferation of call recording and transcription services was supposed to make this easier, not harder. If the words are captured automatically, the argument goes, you are free to give the room your full attention instead of splitting it with a notepad. For me, that has mostly held true. The transcript arrives afterward and does one job: it clarifies a detail or answers a question I did not need to write down because I was too busy listening to the person asking it.

The convenience that cuts both ways

That is not everyone's experience, and it is worth being honest about why. Knowing a transcript will land in your inbox an hour from now is itself a small permission to drift, to answer a message mid-sentence of someone else's point, to multitask on the theory that the record will catch what you miss. The same tool that lets one person watch more closely lets another person watch less. The technology did not decide which one you would be. It just removed the excuse of "I couldn't keep up" and replaced it with a quieter, easier excuse: "I'll check the transcript later."

That is the choice this piece is really about. Not whether transcription is good or bad, but whether you are using it to see more of the person in front of you or to justify seeing less.

What the eyes catch that the record cannot

A transcript is a faithful record of words and a poor record of everything that gave those words their weight. It cannot tell you which answer came a half-second too fast to be considered, or which one carried the kind of hesitation that means someone is protecting a colleague rather than lying to you. It flattens a passionate objection and a bored one into the same sentence structure. Watching the speaker is how you catch the difference between what someone is willing to say and what they actually believe, and that gap is usually where the real information lives. None of this shows up in a searchable document afterward. It only shows up if you were paying attention while it happened.

What changed in 2026

The tools spread fast enough that they stopped being a personal choice and became infrastructure. A 2025 Fellow survey found three out of four professionals already relying on an AI notetaker at work, and Otter.ai, Fireflies, Fathom, and Read.ai ( All video calling software seem to have their own transcription tools plus Google Recorder on Android) now show up in a meeting's participant list as routinely as a human latecomer. That scale is exactly what turned a convenience into a legal question. In re Otter.AI Privacy Litigation, a consolidated federal class action in the Northern District of California, alleges that Otter's notetaker recorded and transcribed participants who never agreed to it, and that the audio was used to train the company's models. The claims run under the federal Wiretap Act and the California Invasion of Privacy Act, which carries statutory damages up to five thousand dollars per violation. Otter's motion to dismiss was argued on May 20, 2026, in front of Judge Eumi Lee, and no ruling has issued as of this writing. Otter's CEO Sam Liang has pushed back publicly, arguing the company is on the right side of history and that privacy law itself needs to catch up to what AI tools now do.

The platforms are not waiting for a ruling to act. Microsoft Teams published a Message Center notice in March announcing that unrecognized third-party meeting bots will be labeled Unverified and routed to a lobby the organizer must explicitly clear. Zoom already requires host or co-host initiation for its own AI Companion, with administrators able to disable it inside third-party meetings. Google Meet's native note-taking feature works the same way, host-initiated and admin-gated. Some institutions have moved even faster on their own. Cornell's IT department automatically blocks Read.ai and Fireflies from its Zoom instance, and the University of Washington, Chapman University, and UC Riverside have taken similar steps without waiting for a court to tell them to.

A transcript remembers every word. It has never once remembered what mattered.

A decision for the person on the call

The old advice to listen instead of transcribing still holds for most conversations that matter. The transcript should be the thing you reach for to confirm a number or a name, not the thing you rely on to have been present at all. If you would not reopen it, it is not doing anything for you except sitting in a vendor's storage. Ask yourself before the call starts whether this conversation needs your eyes on the speaker more than it needs a searchable record, and let that answer, not the default settings on your calendar, decide whether a notetaker joins.

A decision for the person running the call

Analyst briefings, product previews under embargo, and anything tied to a disclosure date sit in a different category, and I say this from the organizing side as often as the receiving one. An embargoed briefing exists because the information has a value that depends on controlled release. A transcription bot in that meeting is an uncontrolled copy of material non-public information sitting on a third-party server, subject to that vendor's retention policy, training practices, and now its own pending litigation. The harm from a leak here is not abstract. It is a scoop, a stock move, or a partner relationship, depending on what the briefing covers. Organizers who have not written a transcription policy for sensitive briefings should disable third-party bot access at the meeting level rather than relying on attendees to self-police, and if a record is genuinely needed, generate it through a tool the organization controls rather than one an individual attendee happened to sign up for.

Shashi Bellamkonda. Principal Research Director, Info-Tech Research Group. Former Adjunct Professor, Georgetown University, Entrepreneur in Residence, Stony Brook University, NY.

Works Cited

"Microsoft, Zoom and Google Tighten Meeting Bot Controls as Otter Case Nears Hearing." UC Today, 6 May 2026.

"The Meeting Bot Nobody Invited Is Now Exhibit A." PYMNTS, 10 June 2026.

"Otter.ai Recording Privacy Class Action." OpenClassActions.com, 14 June 2026.

"Otter.ai Wiretap Lawsuit Explained: AI Notetakers and All-Party Consent." RecordingLaw.com, 20 June 2026.

"Strategies to Block AI Bots from Zoom Sessions." IT@Cornell.