The Meeting Is the Prompt
Claude Meetings puts Claude in a Google Meet call as a visible participant that builds docs and whiteboards while people talk. Its rules on whose tools it may use, and when it leaves, are the interesting part.
On November 10, Anthropic's launch of Whiteboards, Sheets and Data ended with one line about a product that wasn't there yet: "Claude Meetings is coming soon." Two days later it arrived. Introducing Claude Meetings, dated November 12, opens with the whole idea: "Claude can now join your video calls as a visible participant, follow the conversation and start on work during the meeting."
It's a beta, on Team and Enterprise plans, in Google Meet only. I haven't used it. What follows comes from the launch post, the release notes and the Help Center, plus press coverage, marked as such.
The feature list is close to what you'd guess: notes, a transcript, answers, action items. A chat assistant answers to one person, the one typing. A meeting has several people, and they don't all have the same access. Most of the design answers that problem.
| 🧠| Start here: your first meeting with Claude 1. Check you're eligible. You need a Team or Enterprise plan. The beta excludes organizations with zero data retention, a HIPAA-ready configuration or customer-managed encryption keys. Calls must be on Google Meet. 2. Turn it on (owner). Open Organization settings > Products > Claude Meetings (beta). Team has it on for everyone by default. Enterprise has it off until an owner turns it on, for everyone or for custom roles that include it. An owner must also enable Google Calendar for the organization. 3. Check two settings while you're there. The first is "Meetings with external guests". Anthropic's own pages disagree on its Team default, so read the switch itself. The second is "Members' connectors in meetings". It decides whether Claude may use people's connected tools during a call. 4. Pick one internal meeting and tell people. The Help Center's instruction: "make sure everyone on your call is okay with being transcribed." 5. Add Claude. In Claude or Claude Desktop, go to Meetings, connect Google Calendar, select the meeting and click "Add Claude". Leave auto-join off for now. 6. In the call, ask out loud or in chat. Say "Claude" and a question, or type it in the Meet chat. It answers in writing in the chat. Then say "Claude, make a doc" and watch it fill in on the meeting page. 7. Try one tool. Ask it to post a short Slack message. It asks your permission the first time, and if you asked out loud, it confirms who's speaking. Tool use counts toward your plan's usage. 8. Read the meeting page afterwards. It holds the summary, action items, transcript and everything Claude made. Colleagues on the invite get a recap email with the link. Later, ask about the meeting in a regular chat. |
What shipped
The release notes entry for November 12 adds the surfaces: it's "rolling out in beta on Team and Enterprise plans in Claude and Claude Desktop," building "docs, whiteboards, and slides, during the call."
- Where: Google Meet only. The launch post says "Support for other video conferencing platforms is coming soon," with no dates. Mobile isn't mentioned.
- Who can't, yet: ZDR, HIPAA-ready and CMEK organizations (see Start here).
- What it keeps: "a transcript, summary, and screen descriptions, not audio."
- How long it stays: up to eight hours, with a warning before it leaves.
TechCrunch and Fortune reported the announcement on November 10; Anthropic's release notes and Help Center date the rollout to November 12, so a missing Meetings tab may just mean not yet.
According to Fortune, Anthropic presented it at a San Francisco event on the day its IPO roadshow began. Nicholas Lin, who leads knowledge-work product, summed it up on stage as "no more switching tabs." Axios reported, in paraphrase, that Anthropic tested it internally for about two months and that Claude now sits in nearly all of the company's own meetings.
How a meeting runs
You add Claude from your calendar, or turn on auto-join under Settings > Capabilities > Meetings. Auto-join has two scopes: all your meetings, or all except one-on-ones. The Help Center adds one line that matters: "Owners can't set auto-join for you." Claude usually joins without having to ask; if it does ask, the host can let it in. If the host declines, Claude doesn't try that call again unless someone adds it.
In the call it has its own tile, which shows its current state. You address it by name ("Claude, ...") or type in the Meet chat. It doesn't speak. "Claude, make a doc" or "Claude, make a whiteboard" starts an artifact, and the Help Center says all artifact types are supported. Anything Claude makes lands on the meeting page, a claude.ai page for that call with the live transcript, a chat and the artifacts. To pause it, ask it to stop listening, or use the meeting page.
After the call, the meeting page keeps the notes (a summary and action items), the transcript and the artifacts. People from your organization on the invite get a recap email. In a regular chat, you can ask Claude about past meetings you have access to.
That's the difference from a plain notetaker: it builds the work during the call, not a summary after it. The whiteboard fills in while people argue, and the doc exists before anyone has to "send notes around." It builds on the artifacts from the November 10 launch, which I covered in Claude Stopped Handing You Files.
Per Axios, again in paraphrase, Claude largely waits to be told what to do, and Anthropic expects it to become more proactive over time.
The agent in a room of principals
It is visible, by construction. Granola's homepage says "Granola doesn't join your meeting": it transcribes your computer's audio in the background. ChatGPT's record mode records locally on a Mac. Claude Meetings goes the other way. It is a participant with a tile, and the launch post says it "posts a notice in the call's chat" when it joins. A background recorder leaves consent to one attendee; a participant puts it in front of everyone. Checking that everyone is okay with being transcribed is still on you, but nobody can miss that it's there.
Tool authority doesn't pool. Claude "uses each person's connected tools with that person's permission. It asks the first time, and a yes applies going forward." The call as a whole doesn't become a principal with everyone's combined access. If I ask Claude to post in Slack, it posts as me, with my connector and my consent. Owners can switch members' connectors off in meetings entirely.
Spoken requests get re-checked. The documentation is direct about this: "For now, when someone asks out loud, Claude asks each time to confirm who's speaking." Anyone can say "Claude, send that to the client"; audio alone doesn't tell Claude whose tools to use. Typed requests in chat come with an account attached. Spoken ones don't.
Guests make it go quiet. When outsiders join partway through, "including another company's notetaker bot," Claude leaves unless guests are allowed (defaults below). When guests are allowed, the Help Center describes a quieter mode: Claude still takes notes and posts its notice, but it doesn't reply in the Meet chat. Guests' words go into the notes, but guests get no meeting page and no recap. The chat on the meeting page stays inside the organization.
It keeps less raw material. Text and screen descriptions, no audio. That limits what can leak, though a transcript still holds every word said. I found no published retention period for meeting pages or transcripts.
| 💡 | The pattern: an agent that serves several people should be visible to all of them, borrow authority from one named person per action, re-check identity when the channel can't prove it, and say less as the audience gets less trusted. |
The conflict you should check yourself
Anthropic's two main pages disagree on one default. The launch post: "By default, Claude is set to stay out of calls when someone from outside your organization is invited." The Help Center: "On Team plans, Claude can join calls with outside guests by default, and an owner can turn this off. On Enterprise plans, an owner must allow it first." Its admin table repeats it: external guests are "allowed" by default on Team and "internal only" by default on Enterprise.
For Enterprise the two pages agree. For Team they contradict each other. Until Anthropic fixes one of them, Team owners should check the switch itself (Start here, step 3). Each person also has their own "Meetings with external guests" setting.
What it costs
During the beta, the launch post makes most of it free on Team and Enterprise: notes, transcripts and answers in the call, "unless an answer uses one of your connected tools, which counts toward your plan's usage." The docs, whiteboards and other artifacts Claude makes in a call "are free through February 10, 2027." The Help Center adds that using connected tools "counts toward your own usage and can lead to charges for your organization."
Nothing published says what any of it costs after the beta or after February 10.
Where it sits in the field
Claude Meetings is starting on Google Meet, where Gemini already takes notes natively, so Meet customers will compare the two directly. And the export partners Anthropic names, "Gong, Attention, Granola, Fireflies.ai, and Otter," sell meeting capture themselves. That suggests Anthropic cares less about owning the transcript than about the work that follows it.
For developers: nothing to call yet
I found no Meetings API, no MCP server, no webhook and no Claude Code integration. Transcript export is "soon," through those partners. For now, Claude Meetings is a product for people, not a building block for code.
If you want meetings in an agent pipeline today, the route goes the other way: Zoom's and Granola's connectors feed transcripts to Claude (see the table). These don't do what Meetings does (no live artifacts, no in-call actions), but they're scriptable, and Meetings isn't.
What I'd want before turning on auto-join
On paper, the design gets the multi-person questions mostly right. Before I'd let it join every call, I'd want three answers from Anthropic: how long transcripts and meeting pages are kept, which guest default is the real one, and how "confirm who's speaking" holds up once the agent stops waiting to be asked.
