Quick answer
To set ground rules for an AI meeting assistant, agree six things before anyone switches it on: everyone in the call is told and can object; sensitive meetings (HR, legal, health, external guests) are off limits; the assistant only gets the tools it needs; any action it takes, such as sending a message, needs a person’s approval; a named person checks the notes before they are shared; and you know where transcripts are kept and for how long. Write these on one page, check your tool’s settings match, and review them after 30 days.
Key takeaways
- Consent first. Tell everyone before the assistant joins, every time.
- Some meetings stay human-only: disciplinary, legal, health and most client calls.
- Least access: connect only the tools the assistant genuinely needs.
- No silent actions. A person approves anything sent or changed.
- Notes are drafts until a named person checks them.
- One page is enough for a small team, reviewed after a month.
In this guide
- What is an AI meeting assistant?
- Why do you need ground rules?
- Rule 1: Everyone knows, and anyone can say no
- Rule 2: Decide which meetings are off limits
- Rule 3: Limit what the assistant can access
- Rule 4: Actions need a person’s approval
- Rule 5: A person checks the notes before they travel
- Rule 6: Know where notes are kept, and for how long
- Which Claude Meetings settings should you check?
- What should a one-page policy say?
- How should you tell people?
- Frequently asked questions
- The bottom line
- Related posts
This guide follows on from our round-up Gemini 4 Hydrogen: Google’s Fast New Flash Model, Anthropic’s “Slow Down” Pledge and This Week’s AI News, which covers the launch of Claude Meetings, a new beta that lets Claude join Google Meet calls, take notes and build documents while your team talks.
📘 Main article: Read the full round-up in Gemini 4 Hydrogen: Google’s Fast New Flash Model, Anthropic’s “Slow Down” Pledge and This Week’s AI News for Claude Meetings, Gemini 4 Hydrogen and the rest of the week’s AI news from WorldofAI.
An AI that listens to your meetings can save hours of note-taking. It also hears everything said in the room. This guide answers one question: what rules should a small team agree before letting an AI assistant into its calls? The rules work for Claude Meetings and for any similar tool.
What is an AI meeting assistant?
An AI meeting assistant joins your video calls, follows the conversation, and produces notes, action items and sometimes documents during or after the meeting. Anthropic’s Claude Meetings goes a step further: Claude appears as a visible participant, answers questions in the chat, builds documents and whiteboards live, and can use connected tools, such as sending a Slack message, with permission.
According to Anthropic, Claude Meetings is in beta for Team and Enterprise plans, starts with Google Meet, and saves transcripts and screen descriptions rather than audio.
Why do you need ground rules?
Because the risks are about people and data, not technology. Without agreed rules, a few predictable problems appear:
- someone says something in confidence, not realising it is being transcribed;
- a client joins a call and an AI is listening without their knowledge;
- a draft summary with a mistake is emailed to the whole company;
- the assistant has access to tools nobody intended it to use.
In the UK, a transcript that identifies people is personal data, so data protection law applies. The ICO’s guidance is the place to check your obligations. This guide is practical advice, not legal advice.
Rule 1: Everyone knows, and anyone can say no
Tell everyone before the assistant joins, and make it easy for anyone to ask for it to leave. Anthropic’s own help page says to make sure everyone on the call is happy to be transcribed before you add Claude or switch on auto-join.
- Put “AI note-taker joining” in the calendar invite.
- Say it out loud at the start of the call.
- If anyone objects, remove the assistant, with no questions asked.
💡 Tip: Claude posts a notice in the chat when it joins and shows as a participant. That is helpful, but do not rely on people noticing. A sentence at the start of the call is clearer.
Rule 2: Decide which meetings are off limits
Agree in advance which meetings never have an AI assistant. A sensible starting list for a small business:
| Meeting type | AI assistant? | Why |
|---|---|---|
| Team stand-ups and planning | Yes | Low risk, high value from notes and action items |
| Project reviews | Yes, with consent | Useful summaries; check notes before sharing |
| Client or supplier calls | Only with their explicit agreement | Their data and confidentiality are involved |
| One-to-ones and appraisals | No | Personal and sensitive |
| Disciplinary, grievance or HR | No | Highly sensitive personal data |
| Legal, health or financial advice | No | Confidential and regulated |
According to press reports, Claude Meetings will not join calls with external guests unless an admin allows it, and Anthropic lets you exclude one-to-ones from auto-join. Use those settings to back up your rules.
Rule 3: Limit what the assistant can access
Connect only the tools the assistant needs for meetings, and nothing else. If it only needs to write notes, it does not need your email, your files or your payments system.
- List the tools your AI account is connected to.
- Remove any that meetings do not need.
- Check whose permissions the assistant uses. In Claude Meetings, Anthropic says it uses each person’s own connected tools with that person’s permission.
Our guide to writing an AI acceptable use policy for a small team covers access rules for AI tools more broadly.
Rule 4: Actions need a person’s approval
The assistant can suggest and draft, but a person approves anything that leaves the meeting. That includes sending messages, creating tasks for other people, booking time or editing shared documents.
Claude Meetings asks for confirmation the first time it uses someone’s connected tools, and Anthropic says that for now it asks each time to confirm who is speaking when a request is made out loud. Keep your own rule stricter than the tool’s defaults: no message is sent until the person named in the action has read it.
⚠️ Warning: “A yes applies going forward” is convenient but easy to forget. Review the permissions you have granted once a month, and remove any you no longer need. Our guide to adding a kill switch to your AI agents and automations shows how to stop things quickly if something goes wrong.
Rule 5: A person checks the notes before they travel
Name one person per meeting who checks the AI’s notes and action items before they are shared. AI notes are usually good, but they can mix up who said what, miss sarcasm or turn a suggestion into a decision.
- Check names on every action item.
- Check numbers, dates and deadlines.
- Remove anything said off the record.
- Mark the notes as checked before sending.
If several people talk at once, transcripts get less reliable. Our guide to getting an accurate AI transcript when several people are talking has practical fixes.
Rule 6: Know where notes are kept, and for how long
Before you start, find out where transcripts and notes are stored, who can see them, and how to delete them. In Claude Meetings, Anthropic says notes, transcripts and anything Claude made are kept on a meeting page that people from your organisation on the invite can see, and they get a recap email with a link.
- Decide how long you keep transcripts. Many small teams keep the checked summary and delete the full transcript after 30 to 90 days.
- Make sure meeting pages are not shared more widely than the meeting itself.
- Know how to delete a transcript if someone asks.
Which Claude Meetings settings should you check?
Use this as a checklist. Names and menus may change during the beta, so check Anthropic’s help page for the latest.
| Setting | What to check | Suggested choice |
|---|---|---|
| Organisation access | On Team plans it is on by default; Enterprise owners switch it on | Owner decides deliberately, rather than leaving the default |
| Auto-join | Settings → Capabilities → Meetings → Auto-join | Off at first; add Claude to meetings one by one |
| One-to-ones | Auto-join can exclude one-to-ones | Exclude them if you do use auto-join |
| External guests | Reported to require admin permission | Keep blocked unless a client has agreed in writing |
| Connected tools | Which tools Claude can use for each person | Only what meetings need |
| Pause or remove | Ask Claude to stop listening, or use the meeting page | Make sure everyone knows how |
Before your first AI-assisted meeting: copy the policy template below, fill in the names, share it with your team and switch auto-join off until everyone has read it.
What should a one-page policy say?
Copy and adapt this. Keep it to one page so people actually read it.
- Purpose: we use [tool] to take notes and draft action items in internal meetings.
- Consent: the organiser tells everyone before the assistant joins. Anyone can ask for it to leave.
- Off limits: one-to-ones, HR, legal, health and financial advice meetings. Client calls only with written agreement.
- Access: the assistant is connected only to [list tools].
- Actions: nothing is sent, booked or changed without approval from the person responsible.
- Checking: the organiser checks notes before sharing them.
- Storage: summaries kept in [location]; full transcripts deleted after [30/60/90] days.
- Owner and review: [name] owns this policy and reviews it on [date, 30 days from now].
How should you tell people?
A short, calm message works best. For example:
“From next week, we will trial Claude as a note-taker in our internal team meetings. It will show as a participant and post a notice when it joins. The organiser will always mention it at the start, and anyone can ask for it to leave. It will not join one-to-ones, HR or client calls. Notes are checked by the organiser before they are shared, and full transcripts are deleted after 60 days. The one-page policy is attached. Questions or concerns? Talk to [name].”
Frequently asked questions
Do I need everyone’s consent to use an AI note-taker?
You should always tell everyone and let them object. Whether you also need formal consent depends on your situation and the law where you work, so check guidance such as the ICO’s for the UK.
Can Claude Meetings join calls with clients?
Press reports say it will not join calls with external guests unless an admin allows it. Even if allowed, ask the client first.
Does Claude Meetings record audio?
Anthropic says it saves a transcript, a summary and descriptions of what was on screen, not audio.
Which plans include Claude Meetings?
It is rolling out in beta on Claude Team and Enterprise plans, starting with Google Meet. Some organisations, such as those with zero data retention, cannot use it during the beta.
Is Claude Meetings free?
During the beta, Anthropic says notes, transcripts and most in-call answers are free on Team and Enterprise, and documents and whiteboards made in calls are free until 10 February 2027. Using connected tools counts towards your plan’s usage.
What if the AI notes are wrong?
Treat them as a draft. The person checking them corrects mistakes before sharing, which is why that role matters.
The bottom line
An AI meeting assistant is most useful when everyone trusts it. That trust comes from simple, visible rules: people know when it is there, it stays out of sensitive meetings, it cannot act on its own, and a person checks what it writes. Agree those rules on one page before the first call, and you get the time savings without the awkward surprises.
Related posts
- Gemini 4 Hydrogen: Google’s Fast New Flash Model, Anthropic’s “Slow Down” Pledge and This Week’s AI News
- How to Write an AI Acceptable Use Policy for a Small Team
- How to Get an Accurate AI Transcript When Several People Are Talking
