Meeting Notes With AI: Consent, Filing and What Swiss Law Allows

Key Takeaways

• AI meeting notes come from a recording, a transcript and a summary, which means the tool has to listen in.
• Recording a private conversation without everyone's consent is a criminal offence, even if you're taking part in it.
• Announce the recording in the invitation, ask again at the start and accept a no.
• The checked notes belong in the file they relate to; delete the audio and raw transcript once they're approved.
• Transcripts are for the minutes, not for monitoring staff.

AI can turn a meeting into minutes with decisions and action points within a few minutes, provided it's allowed to record or transcribe the conversation. In Switzerland that means everyone taking part knows beforehand that it's being recorded and why, and agrees. Recording a private conversation without that consent is a criminal offence, even if you're one of the people in it.

How does AI produce meeting notes?

In three steps: recording, transcript and summary. The tool records the audio, whether in a video call, through a device in the meeting room or with an app on the phone. Speech recognition turns it into a transcript showing who said what. A language model then summarises it, usually into decisions and open points with owners and dates, and puts it into your template. How well this works depends on sound quality, jargon and language. So test it with real meetings before you commit, especially if people speak Swiss German.

Can I simply record a meeting?

No, only with the consent of everyone taking part. Under Article 179ter of the Swiss Criminal Code, anyone who takes part in a private conversation and records it without the others' consent faces up to a year's imprisonment or a fine. Keeping, analysing or passing on a recording made that way is also an offence. Listening in on or recording a conversation you're not part of falls under Article 179bis, with a penalty of up to three years.

The exception for business phone calls (Article 179quinquies) is narrow. It covers orders, contracts, reservations and similar business transactions, and such recordings may only be used as evidence. It doesn't help with meeting notes. The FDPIC stresses that consent should be as clear as possible about the recording and its purpose, and that it has to be obtained before recording starts, so people have the chance to object.

What does data protection law add?

A transcript contains personal data, so the principles of the FADP apply: transparency, a recognisable purpose, proportionality and deletion once the data is no longer needed (Art. 6 FADP). In practice:

• Inform people: say who's responsible, what the notes are for and who receives them (Art. 19 FADP). If the provider processes the data abroad, name the country too.
• Stick to the purpose: a transcript made for the minutes isn't material for performance reviews. Article 26 of Ordinance 3 to the Employment Act (ArGV 3) prohibits monitoring systems meant to watch employees' behaviour at work. A note-taking tool should never turn into one.
• Check the provider: it processes the data on your behalf. Find out where it's stored, which subcontractors are involved and whether your conversations are used for training (Art. 9 FADP).
• Set deletion periods: you usually need the audio and raw transcript only until the minutes are approved.
• Be ready for access requests: anyone can ask whether and which data you process about them (Art. 25 FADP). Whatever you keep, you need to be able to find.

Conversations with a law firm's clients are also covered by professional secrecy, which we look at in AI in law firms. The general framework is explained in AI and the Swiss FADP. As of October 2026. This article isn't legal advice.

Which conversations suit AI meeting notes?

Meetings that get minuted anyway and where nobody shares anything personal. Good candidates are:

• Weekly team and status meetings with a list of action points.
• Project and site meetings, where decisions and dates are recorded for several parties.
• Internal workshops that should end with a summary and next steps.

Be more careful with performance reviews, conversations about health or conflicts, confidential negotiations and client meetings under professional secrecy. There, even asking to record is delicate, because hardly anyone feels free to say no. Handwritten notes that you tidy up with AI afterwards are often the better choice.

How do you get consent in everyday practice?

With a short routine that runs the same way at every meeting:

1. Announce it in the invitation.Say in the invitation that an AI tool will record and produce minutes, who will receive them and when the recording will be deleted.
2. Ask at the start.Before you begin, ask explicitly whether everyone agrees, and note the answer in the minutes. This matters most with external guests.
3. Accept a no.Nobody has to explain why they'd rather not. Someone takes notes by hand and the tool stays off. Consent only counts if it's given freely.
4. Record visibly.Use tools that show everyone a recording is running, and pause for confidential points.
5. Do the same on the phone.Ask before the recording starts. The exception for orders and reservations doesn't cover call notes.

What should you look for in a tool?

Mainly where the data is kept, when it's deleted and whether the minutes end up where your team works. In detail:

• Storage and contract: where the audio, transcript and minutes are kept, and a written commitment that your conversations won't train any model. There's more in AI tools and where your data goes.
• Deletion: whether the audio and transcript are deleted automatically after a set period.
• Visibility: whether everyone in the meeting can see that a recording is running.
• Filing: whether the minutes can be moved into your client file or project folder.
• Language: how well the tool copes with Swiss German, jargon and names.
• Administration: whether you can set centrally who uses the tool and with which settings.

Where should the finished notes go?

Wherever the matter they're about is filed: the client file, the project folder, the property file. Not in the note-taking tool's cloud, where nobody can find them a few weeks later. Before they're filed, the person responsible reads the summary and corrects it, because AI sometimes mixes up speakers, misses a no or turns an idea into a decision. The minutes then get the same access rights as the folder they sit in. Send the approved minutes to everyone who took part, so they can point out mistakes; that's part of being open about the recording too.

An example: a property management firm goes through its open damage cases every Monday. The invitation mentions the note-taking tool, and at the start everyone confirms they agree. After the meeting, the team lead checks the draft, fixes a date the tool misheard and files the action points under each property. The audio is deleted as soon as the minutes are approved.

Which tools your team may use for this is best settled in your AI usage policy. How minutes, reports and other paperwork can take less of your time overall is shown on our documentation page.