AI Note-Takers Now Join Client Calls as Participants. A Solo Business Should Ask First, Every Time

Close-up of a studio microphone lit by vivid pink and blue neon light

5 min read

In roughly a dozen US states, including California, Florida, Illinois, Pennsylvania and Washington, recording a private conversation generally needs the consent of everyone in it, not just one person. The Digital Media Law Project’s state-by-state guide is a good place to check your own state and your clients’. Keep that in mind, because AI note-takers have just taken a step that makes the question harder to ignore.

On November 12, Anthropic put Claude Meetings into beta for Team and Enterprise accounts. Claude can join supported video calls as a visible participant, answer in writing during the meeting, and build documents from what is said, with notes and transcripts kept on the meeting page. Other tools already do versions of this. But an assistant that sits in the call, listens, and acts while people are still talking is a different thing from a recorder you switch on afterwards.

My position is simple: a solo business should never let an AI join a client call without asking the client first, every time, out loud or in writing. Not because the law always requires it, but because your reputation does.

A visible bot is not the same as permission

The common defense is that the AI shows up as a named participant, so everyone can see it. That is better than a hidden recorder. It is not consent.

Think about the client’s side of the call. They join, they see your face and a tile that says something like “Claude” or “Notetaker”. Are they going to stop the meeting and ask what it records, where the transcript goes, and who trains on it? Most will not. They will feel slightly watched, say less than they meant to, and wonder later whether you asked anyone else. Silence is not agreement. It is often politeness.

For a big company with a legal team and a published recording policy, a visible bot may be a reasonable default. For a one-person business, your relationship with each client is the business. One uncomfortable call can cost more than every note the AI will ever take.

The useful part is real, which is why the habit matters

I am not arguing against AI note-takers. They are some of the most useful AI tools a solo consultant can use. We compared the main options in Fathom vs Otter vs Fireflies vs Granola, and the time savings are genuine. An assistant that drafts the follow-up email before you hang up is even more useful.

That is the point. The tools are good enough that you will want them on every call. So the habit you build now, while they are new, is the habit you will have for years. Build the respectful one.

What asking actually looks like

Asking does not need to be awkward. Three layers cover almost every situation:

  1. In the calendar invite: “I use an AI assistant to take notes on calls so I can focus on you. It joins as a visible participant. If you would rather it did not, just reply and I will leave it off.” This gives people a no-pressure way to say no.
  2. At the start of the call: “Before we start, is it still OK for my note-taker to join?” Five seconds. Wait for a yes.
  3. In your contract or terms: one clause that says you may use AI tools for notes and drafting, how long you keep transcripts, and that clients can opt out. Our guide on writing an AI disclosure clause has a template you can adapt in an hour.

When a client says no, turn it off without fuss and take notes by hand. That moment builds more trust than any summary.

The calls where you should not bring an AI at all

Some calls deserve a hard rule, even with permission:

  • Calls about a client’s health, money troubles, legal disputes or staff problems. The more sensitive the topic, the less you want a transcript sitting anywhere.
  • Calls where the client is clearly upset. A complaint call is not the time to add a participant.
  • Calls with a third party you have not met, such as your client’s lawyer or accountant. You cannot assume their consent.
  • Calls where you have promised confidentiality in writing, unless your agreement explicitly covers AI tools.

Check the plan before you check the feature

Claude Meetings is in beta for Team and Enterprise accounts, which are business plans with admin controls. Before you turn on any in-call assistant, read three things in its settings or help pages: who can see the transcript, how long it is kept, and whether it is used to train AI models. If you cannot find clear answers, that is your answer.

Our earlier look at Anthropic’s new office tools covers what each one does. This piece is about the one rule that should sit on top of all of them.

The standard to set for 2027

AI assistants will keep moving from “record and summarize later” towards “take part and act now”. Each step makes them more useful and more present in the room. The businesses that come out of this with stronger client relationships will be the ones that treated consent as a courtesy rather than a checkbox.

So here is the standard I would set: ask in the invite, confirm at the start, honor every no, and keep the sensitive calls human. It costs you about ten seconds per meeting. It tells every client that you value them more than the tool.

Do you ask before your note-taker joins, or do you rely on the bot being visible? Make your case in the comments.

Related reading

Leave a Comment

Scroll to Top