An AI note taker can transcribe a meeting and turn it into a summary, a list of decisions and a list of action items a few minutes after the call ends. It does not replace a review: names, numbers and decisions still need to be checked by someone who was in the room. And because the tool processes people's voices, you need to tell participants and know where the data goes.

Why meeting notes get lost

In a small business, whoever runs the meeting jots down a few lines, plans to clean them up that evening, then moves on. Three days later, the details are gone and the follow-ups have started with nothing in writing. The problem is not the writing, it is the delay, and this is what AI shortens: a first draft is ready minutes after the meeting.

What AI does well, and what it gets wrong

What it does well

  • Transcription: speech becomes time-stamped text. On a video call, where everyone has their own microphone, the tool usually knows who is speaking.
  • Structured summaries: a language model pulls out the topics covered, the decisions and the next steps.
  • Search: a transcribed meeting can be found months later by keyword.

What it gets wrong

  • Proper nouns: customers, products, suppliers, internal acronyms. "Whitfield" can come out as "Wheatfield".
  • Numbers: amounts, quantities, dates. Mixing up "fifteen" and "fifty" is easy to miss on a quick read.
  • Crosstalk: when two people talk at once, or a whole room shares one microphone, speaker attribution becomes approximate.
  • Mixed languages: some tools handle only one language per meeting. Google Meet's note taking is one of them.
  • The line between a proposal and a decision: this is the most important error. An idea floated ("we could push the delivery back") can show up as a decision, and a decision nobody stated out loud can be missing.
  • Owners: an action item that says "send the quote" with no name and no date will be done by no one.

The 7-step method

1. Tell participants

Mention the transcription in the invitation, then repeat it at the start of the meeting: what it is for, who will receive the notes, and how to object (see the GDPR section below).

2. Set up the tool before the meeting

Check the transcription language, where the summary is saved and who receives it: some tools can send it to every invitee, external guests included.

3. Speak for the transcript

A few habits make a clear difference:

  • one microphone per person when possible, and one person speaking at a time;
  • customer names, product names and numbers said clearly;
  • decisions stated out loud: "Decision: we go with carrier B.";
  • action items with a name and a date: "Jake follows up with the printer before Tuesday."

4. Generate the summary with a fixed structure

Ask for a summary that always follows the same headings: context, topics discussed, decisions, action items (what, who, by when). If you are starting from a raw transcript, our meeting minutes template gives you a ready-made structure.

5. Review and correct

Set aside a few minutes right after the meeting. Check first:

  • the spelling of people, customer and product names;
  • every number, amount and date, going back to the transcript when in doubt;
  • that each decision was actually made, not just discussed;
  • that each action item has an owner and a due date;
  • that no sensitive information appears in the summary without a reason.

6. Share the summary, not the transcript

Nobody reads forty minutes of dialogue, and transcripts often contain remarks that were never meant to circulate.

7. Follow up on action items

Copy the action items into your task tool and review the open ones at the start of the next meeting.

The main types of AI meeting tools

Tools fall into four groups. The choice depends mostly on where your meetings take place and where you want the data stored.

Type Examples How it works What to check
Built into your video tool Teams, Google Meet, Zoom AI is turned on inside the video tool you already use License required, storage location, who gets the summary
Bots that join the meeting Noota, Leexi, tl;dv, Otter, Fireflies A bot joins the call as a participant, records, then transcribes Hosting, subprocessors, informing external participants
Bot-free note-taking apps Granola The app captures audio from your computer or phone without joining the call Informing participants, hosting location
Workspaces with built-in video Omeon Video calls, transcription and notes live in the same tool Coverage of meetings held outside the tool

Built-in features in Teams, Meet and Zoom

Microsoft Teams offers a meeting recap (recording, transcript, notes, tasks). AI-generated notes and tasks, called intelligent recap, require a Teams Premium or Microsoft 365 Copilot license, and transcription must be turned on (Microsoft documentation).

Google Meet offers "Take notes for me", powered by Gemini. Notes are saved to a Google Doc in the organizer's Drive, and the organizer chooses who they are shared with. The feature requires an eligible Google Workspace edition or Google AI plan and supports one language per meeting (Google Meet Help).

Zoom produces meeting summaries with next steps through its AI assistant, included in most paid plans (Zoom page).

Nothing to install, but your data follows the vendor's hosting, and a higher-tier license is often required.

Bots that join the meeting

These tools work with Teams, Meet and Zoom: a bot joins the call, appears in the participant list, records, then transcribes. Hosting varies by vendor:

  • Noota (France): European customers' data is stored in Belgium and the Netherlands on Google Cloud, with an option hosted at Scaleway in Paris (Noota security page).
  • Leexi (Belgium): recordings and transcripts hosted on AWS in Paris, with an OpenAI language model run through Azure in France (Leexi compliance FAQ).
  • tl;dv: data mainly stored and processed in the European Economic Area, with a choice of AI processing in Europe or the United States (tl;dv article).
  • Otter: data stored in the United States on AWS, with transfers covered by the EU-U.S. Data Privacy Framework (Otter privacy policy).
  • Fireflies: data stored and processed in the United States by default; the Enterprise plan allows storage in the EU, while processing stays in the United States (Fireflies documentation).

Bot-free note-taking apps

Granola takes a different approach: no bot joins the call. The app, available on Mac, Windows, iPhone and Android, captures your computer's audio and transcribes it live, then adds a summary and next steps to your own notes. It works with any video tool, and even in person with your phone. According to its security page, audio is not stored, notes are hosted in the United States on AWS, transcription goes through providers such as Deepgram and AssemblyAI, and summaries through OpenAI or Anthropic models. A data processing agreement (DPA) is available on request.

The catch: since no bot shows up in the participant list, other people cannot see that a transcription is running. It is up to you to tell them.

Workspaces with built-in video

Another approach is to hold the meeting in the tool that already holds your notes, tasks and contacts, which is what Omeon does. Video calls run on our own infrastructure, hosted in Europe, with no third-party bot to invite. The host starts the transcription during the call and can stop it at any time. By default, meetings in French are transcribed with Voxtral, Mistral's model, and meetings in English with OpenAI's Whisper. Audio is never sent to an AI provider your organization has excluded: if it excludes OpenAI, Voxtral is used for every language. When the meeting ends, a summary is generated under four headings: summary, key points, decisions, next steps. The meeting record is private by default, and external participants cannot access the transcript after the call.

The limit: today, this approach covers meetings held in the tool, not those you are invited to on Teams, Meet or Zoom. Omeon plans to offer, from mid-November 2026, transcription and summaries of these external meetings from its desktop apps (Windows and macOS).

GDPR and privacy: what to check

Voices and what people say in a meeting are personal data, and so are the transcript and the summary. If you or your participants are in the European Union, the GDPR applies, even to a four-person internal meeting.

Tell participants

Under the GDPR (Article 13), people whose data you collect must be told why it is collected, who receives it, how long it is kept and what rights they have, including the right to object where it applies (GDPR text on EUR-Lex). In practice: a line in the invitation, then a reminder out loud at the start.

Informing people is not the same as collecting their consent: whether you need consent depends on the legal basis of the processing. Recording a conversation is also covered by rules outside data protection, and those differ from one country to another. Check the recording-consent rules that apply in your country, and in your participants' countries when the call crosses borders.

The simple rule: if a participant objects, stop the transcription and take notes by hand.

Know where the audio and transcript go

Ask the vendor three questions:

  1. Where are the audio, transcript and summary stored? Country and hosting provider.
  2. Who are the subprocessors? In particular the transcription provider and the model that writes the summary, which are often not the vendor itself.
  3. Is data transferred outside the European Union? Transfers are allowed, but they must be covered by one of the safeguards in Chapter V of the GDPR.

Also ask whether your data trains models, and check the data processing agreement (GDPR Article 28).

Limit how long you keep it

The GDPR does not allow personal data to be kept longer than needed for its purpose (Article 5). Keep the meeting notes as long as the project needs them, delete the raw transcript once the notes are approved, and set a short retention period if the tool keeps the audio.

Keep sensitive meetings out of it

One-on-ones, disciplinary matters, health issues: these meetings involve sensitive information, and health data is one of the categories the GDPR protects more strictly (Article 9). When in doubt, do not transcribe them.

Summary

AI saves most of the writing time if you keep three habits: tell participants, check names, numbers and decisions, and know where the data is stored. To choose a tool, start from where your meetings happen, or use a workspace like Omeon if you want meetings, transcripts and tasks in one place, hosted in Europe with the database in France.

FAQ

Can AI write meeting notes on its own?

It produces a usable first draft, not a document to send without review. Errors mostly involve proper nouns, numbers and the difference between a decision and a proposal.

Do I need participants' consent to transcribe a meeting?

At a minimum, tell them before the meeting and at the start: purpose, recipients, right to object. Whether consent is required depends on the legal basis for the processing and on the recording rules in your country. If someone objects, stop the transcription.

Can AI take notes in an in-person meeting?

Yes, with a recording app and the same rules for informing people. With one microphone for the room, put it in the middle of the table and avoid talking over each other.