Your Notetaker Just Became a Third Party

Twenty-three pages on one question

On August 13, a federal judge in the Northern District of California issued a written order in a case that most sales teams have never heard of and all of them are affected by. The question in front of Judge Eumi K. Lee was narrow and strange: when an AI notetaker joins a meeting, is it the host’s tool, or is it somebody else in the room?

Otter.ai argued the first thing. If the bot is just the host’s recording device, then the host is a party to the call, the host consented, and that is the end of it. The court did not accept that at this stage. Judge Lee found that because plaintiffs plausibly allege Otter independently collects, retains, and uses communications for its own commercial purposes, they have sufficiently alleged that Otter is a third-party eavesdropper under section 631 of the California Penal Code. Federal Wiretap Act claims and Illinois biometric claims over voiceprints also survived. Several other counts were dismissed with leave to amend.

Two things need saying immediately, because this is exactly the sort of ruling that gets repeated badly. Nobody has been found liable. A motion to dismiss decides whether a complaint is allowed to proceed, not whether the allegations are true, and the case bundles putative class suits first filed in August and September 2025, which now head toward discovery. This is the beginning of the argument, not the verdict.

But the distinction the court drew is the interesting part, and it does not depend on how the case ends. The line was not drawn at recording. It was drawn at what the service does with the recording afterward for its own benefit. That line runs straight through the middle of the software category your team turned on last year without asking anyone.

One in three meetings, and a quarter never asked

The legal argument would be academic if these things were rare. They are not. In a survey of 500 employed US adults conducted on July 8, 2026, 33.4% said an AI notetaker or transcription bot had been present in their work meetings, and 22.4% could not say whether they had been recorded at all.

Among the people who knew a bot had been there, only 34.7% said they were always asked first. A further 36.5% were asked sometimes, 25.1% were never asked and simply watched it appear in the participant list, and 3.6% found out afterward.

Read that from the other side of the table. One in four people who sat in a meeting with a bot never agreed to it. Some of those people were your prospects.

What this does to a sales call

Set the law aside for a moment, because the commercial cost arrives long before any court does.

A discovery call works on candor. You are trying to find out what is actually broken, who actually decides, what the real budget is, and which internal fight your champion is losing. None of that is in the RFP. All of it comes out sideways, in the part of the conversation where somebody says something slightly indiscreet about their own company.

People do not say those things into a transcript they did not agree to and cannot see. They give you the sanitized version, the one that would survive being forwarded. You still get a meeting summary. It is just a summary of a meeting where nobody told you anything.

The irony is worth sitting with. The tool exists so you stop missing things, and the first thing it changes is what people are willing to say in front of it.

The asymmetry nobody mentions

Your prospect cannot see your notetaker’s retention policy, who at your company can search the transcript, or whether the vendor trains on the audio. You often cannot see that either. When the answer to all three is a shrug, asking permission stops being a formality.

A transcript is not a contact record

Here is where this quietly becomes a data problem rather than an etiquette problem.

Watch what happens at the end of a recorded call. Someone reads out a cell number. Someone spells an email address twice, badly. The bot writes down what it thought it heard. Later, a rep goes back into the transcript to dig those details out, and whatever ends up in the CRM came from an automated guess at spoken digits, sitting inside a document whose legal status is now genuinely contested.

That is the worst contact record you could design on purpose. It is unverified, it degrades every time it is repeated aloud, and its provenance is a recording the other person may not have agreed to.

The fix is dull and it works: take the contact exchange out of the transcript entirely. A digital card shared as a plain link in the meeting chat costs the other person one tap, needs no app on their side, and puts your details into their phone as you actually spell them rather than as a transcription model heard them. In person, the same card shares by QR code, NFC tap or AirDrop, and it can live in Apple Wallet or Google Wallet so it is available when the phone is at three percent in a conference hall. If you run discovery calls for a living, get the HeyDrop app and stop mining your own recordings for phone numbers.

It works in reverse too, for anyone still coming home from events with a stack of paper. An AI contact scanner that reads cards and conference badges into structured contacts gives you a record with a known origin and date, which is a different category of thing from a name that surfaced in a transcript.

The team version of this problem

One rep handling this well is a personal habit. Fifteen reps handling it fifteen ways is an operations problem, and it shows up on the day somebody asks where a contact came from.

If every rep shares the same way, from cards created and controlled centrally, then the answer to that question is always the same and always available. Joiners and leavers get handled in the admin panel instead of inside a dozen personal phones, branding stays consistent across the team, and what the team collects comes back in one shape, exporting to CSV or Excel as a CRM-ready record. That is a very different position from reconstructing your pipeline out of meeting summaries. If your team’s contact data currently lives wherever each rep left it, set your team up on HeyDrop.

Five things to change before your next call

Ask, out loud, every time. Not in the calendar invite nobody reads. In the first thirty seconds, as a sentence, with a pause afterward. It costs you nothing and it is the single fact a plaintiff’s lawyer would care about most.

Find out what your vendor keeps and why. Retention period, who can search it, and whether audio or transcripts are used to improve the product. The distinction the court drew was about exactly this.

Know where your prospects sit. A minority of states require consent from every party to a recorded conversation, and California is one of them. On a call with attendees in several states, the strictest one governs your behavior.

Stop treating the transcript as a system of record. It is a memory aid. Contact details, commitments and next steps belong in the CRM, entered deliberately.

Keep one meeting a quarter with no bot in it. The one where you need the truth. You will notice the difference in the first ten minutes.

The part that does not depend on the ruling

Whatever happens to this case in discovery, the underlying shift has already occurred. A thing that listens to your customer conversations, keeps them, and uses them for its own purposes is now a participant in your sales process, whether or not a court eventually calls it one.

You can work with that. It just means the trust you used to get for free from being in a room with someone now has to be earned explicitly, in the first thirty seconds, by asking. The teams that do it will get the honest version of the meeting. Everyone else will get an excellent summary of a conversation that never really happened.

This article and its accompanying image were generated with the assistance of AI.