# How do AI meeting note-takers work?

Date: 2024-03-25 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/ai-meeting-note-takers-work/

**Answer:** An AI note-taker records or listens to a meeting, turns speech into a written transcript, and then uses a language model to summarize that transcript into key points, decisions and action items. The summary is only as good as the transcript and the model's reading of it, so it needs a quick human check, and everyone in the meeting should know it is being recorded.

AI meeting summaries have moved from novelty to a standard feature in a year. Zoom includes its AI Companion at no extra cost for paid accounts, and after a meeting the host can receive an automatic summary to share with attendees and anyone who missed it. Microsoft's Teams Premium offers "intelligent recap," with AI-generated notes, suggested tasks and personalized highlights. Standalone services such as Otter offer the same kind of help outside any one meeting platform.

They are genuinely useful, especially for people who can't attend or who want to listen instead of typing. But they work in a particular way, and knowing how helps you catch their mistakes.

## From voice to summary in three steps

Almost every note-taker follows the same pipeline.

1. **Capture.** The tool records the meeting audio, either built into the meeting platform or through a bot that joins the call as a participant.
2. **Transcription.** Speech recognition software turns the audio into text, usually labelling who spoke when. Microsoft, for example, uses the transcript to split a Teams recording into chapters and to show a timeline of when each person spoke.
3. **Summary.** A large language model, the same kind of AI behind ChatGPT, reads the transcript and writes a condensed version: main topics, decisions and action items. Microsoft says Teams Premium uses OpenAI's GPT models for this step.

Each step can introduce errors. Microphones miss quiet speakers. Transcription stumbles on names, accents, jargon and people talking over each other. And the language model may give a passing remark the weight of a decision, merge two people's views, or assign a task to the wrong person, because it only sees words, not tone or the room.

## A five-minute check before you share

Treat the AI summary as a first draft written by a new assistant. This routine catches most problems:

| Check | What to look for | Fix |
|---|---|---|
| Decisions | Is each listed decision something the group actually agreed, not just discussed? | Delete or reword "decisions" that were only ideas |
| Action items | Right person, right task, right deadline? | Confirm owners; add dates if none were set |
| Numbers and names | Budgets, dates, product and client names | Compare with the transcript or your notes |
| Missing pieces | Was anything important said quietly, off-topic or by a remote participant? | Add it yourself |
| Sensitive content | Personal, HR, health or confidential client details | Remove before sharing widely |

If your tool lets you ask questions about the meeting, use the transcript to verify: ask "Who agreed to send the revised quote, and what exactly did they say?" and read the original passage it points to. Zoom's AI Companion, for instance, lets attendees ask a side panel what they missed during the meeting if the host enables it.

> **Try it:** In your next recorded meeting, take your own brief notes of decisions and action items. Afterward, compare them with the AI summary line by line. Count what the AI got right, what it missed and what it invented. Three meetings will tell you how far you can trust your tool.

## Who controls the data

Ask three questions about any note-taker before you rely on it. Who can turn it on? Zoom's AI Companion is off by default, and account owners and administrators decide whether to enable it. Where do recordings and transcripts go, and for how long? And is meeting content used to train AI models? Zoom says it does not use customer audio, video, chat or similar content to train its own or third-party models. For other tools, read the vendor's privacy policy and your organization's settings rather than assuming.

## What consent do you need before recording?

The rules depend on where you and the other participants are, and this is general information, not legal advice. The safe practice everywhere is simple: tell everyone at the start that the meeting will be recorded and summarized by AI, explain what the recording is for and who will see it, and give people a chance to object or leave before you start. Many platforms show a recording notice, but a spoken announcement is clearer, especially when a bot joins with an unfamiliar name.

In Canada, privacy laws generally expect organizations to collect personal information only for a clear purpose, with meaningful consent, and to protect it. A recording of people's voices and what they said counts as personal information. In Québec, Law 25 has strengthened these expectations in stages since 2022. Québec's privacy regulator, the Commission d'accès à l'information, lists new consent rules among the obligations that took effect in September 2023, along with duties to be transparent, to destroy or anonymize information once its purpose is met, and to offer the highest privacy setting by default in technology products offered to the public. Its checklist also suggests telling people before using technology that can identify, locate or profile them.

In practice, for a Québec business using an AI note-taker, that means: announce the recording and the AI summary, say why you are making it, keep it only as long as needed, limit who can access it, and check where the vendor stores and processes the data. If participants are clients or members of the public, or if the meeting covers health, HR or other sensitive topics, ask whether a recording is needed at all. When in doubt, the Commission's resources and a privacy lawyer are the right next step.

## Sources

- Zoom introduces Zoom AI Companion — available at no additional cost with paid Zoom user accounts: https://web.archive.org/web/20240225055016/https://news.zoom.us/zoom-ai-companion/
- Microsoft Teams Premium: Cut costs and add AI-powered productivity: https://web.archive.org/web/20231229002120/https://www.microsoft.com/en-us/microsoft-365/blog/2023/02/01/microsoft-teams-premium-cut-costs-and-add-ai-powered-productivity/
- Vers la conformité à la Loi sur le privé — Aide-mémoire : nouvelles responsabilités des entreprises, pistes d'action et bonnes pratiques: https://www.cai.gouv.qc.ca/documents/CAI_Guide_obligations_entreprises_vf.pdf