To take meeting notes with AI, tell everyone you are recording and get a yes, record the conversation with a note taker that fits the setting (a bot or desktop app for online calls, a phone on the table for a room), then read the summary against the transcript and fix names, decisions and owners before you share it.
An AI note taker does three jobs: it records, it transcribes, and it summarizes. The method around it matters more than the tool. Most bad AI notes come from a recording nobody agreed to, a phone in the wrong place, or a summary that was forwarded without being read.
This guide is a method you can use with any tool. It covers what to do before, during and after the meeting, and where the AI needs your help.
What does an AI note taker actually do?
It captures audio, converts speech to text, splits the text by speaker where it can, and then asks a language model to write a summary. A typical summary has an overview, key points, decisions and action items.
What it does not do is just as important:
- It does not know anything that was not said aloud. If the decision was made in a chat thread, it is not in the notes.
- It does not know who people are unless you tell it. Most tools start with “Speaker 1” and “Speaker 2”.
- It can mishear names, product terms and numbers, and it can attach a sentence to the wrong person.
- It does not decide what matters to your team. It guesses from the words.
So the notes are a draft with a full record behind it. Your job shifts from typing to checking.
Which kind of AI note taker do you need?
There are two families, and they solve different problems.
| Joins or captures an online call | Records the room on a phone | |
|---|---|---|
| How it hears | A bot joins the call as a participant, or an app captures your computer’s audio | The device microphone picks up the people around it |
| Works for | Zoom, Google Meet, Microsoft Teams calls | In-person meetings, 1:1s, site visits, hallway conversations |
| What others see | A bot is listed as a participant; computer audio capture shows nothing in the call | A phone on the table |
| Weak spot | Does nothing for a meeting in a room | Does not capture the far side of a video call well |
Examples of the first family, described from their own pages:
- Otter’s Notetaker “automatically joins your Zoom, Google Meet, and Microsoft Teams meetings” and is visible in the meeting as a guest participant.
- Fireflies joins at the start time and “appears as Fireflies.ai Notetaker”; it also has a desktop option that records without the bot.
- Granola “uses your system audio to capture transcription” on your computer and says it does not record or save the audio.
If most of your meetings are online calls, one of these is the better choice, and a phone recorder is the wrong tool.
The second family is a recorder app on a phone. Apple’s own Notes app can record audio and show a transcript on iPhone 12 and later in supported languages. Nuvi, the AI note taker for iPhone, iPad and Apple Watch by Ege Beşe, is also in this family: it records the room, separates speakers and writes a summary with decisions and action items. It has no meeting bot and does not join calls. There is more on that trade-off in AI note taker without a bot.
What should you do before the meeting?
1. Tell people and get a yes. Recording others may require their consent depending on where you are, and in some places it requires everyone’s consent. Tell people before you record, say what the notes are for, and wait for agreement. Then say it again once the recording is running, so the agreement is on the recording. The details by region are in is it legal to record a meeting.
2. Check your organization’s rules. Some employers and clients forbid recording or restrict which tools may process meeting audio. A lawful recording can still break a policy.
3. Teach the tool your vocabulary. If the tool has a custom vocabulary, add the names, product names and acronyms that will come up. This is the cheapest accuracy gain available.
4. Place the microphone. For a room, one phone in the middle of the table beats a phone in front of you. Keep it away from laptop fans, projectors and rustling paper.
What should you do during the meeting?
Let the tool take the words and spend your attention on the conversation. Write down only what the recording cannot contain:
- Who is in the room, in the order they first speak.
- What vague references mean (“the Q3 thing” is the pricing change).
- Anything agreed before the recording started.
Two speaking habits improve the result more than any setting:
- Say decisions as decisions. “So we are going with option B” gives the summary a sentence to quote.
- Say owners and dates aloud. “Maya sends the draft by Friday” becomes an action item with an owner and a due date. An action item nobody voiced will not appear.
How do you check an AI summary?
Checking takes a few minutes and is the step most people skip.
- Read the decisions first. For each one, open the transcript at that point and confirm it was a decision and not a suggestion someone floated.
- Check every action item. Is the owner right? Was the date said, or did the model infer it?
- Scan for names and numbers. These are where speech models fail most visibly. Fix them once in a tool that applies the correction everywhere.
- Check the speakers. If the transcript swapped two people, an action item may have the wrong owner. How this happens is explained in speaker diarization explained.
- Listen where it matters. If a sentence carries legal, financial or personal weight, play that part of the audio.
If the summary and the transcript disagree, the transcript wins. If the transcript and the audio disagree, the audio wins.
What should you do after the meeting?
Send a short message, not the whole transcript: the decisions, the action items with owners and dates, and open questions. People read five lines. The difference between these formats is covered in meeting minutes vs notes vs action items.
Then move your own tasks into the place you track work, and delete recordings you no longer need. A recording of other people is something you are responsible for.
How Nuvi fits this method
Nuvi is built for the in-room half of the problem, and it is honest to say where it stops.
- What it does. It records on iPhone, iPad or Apple Watch, with the screen locked if you like. “Room” microphone mode is meant for meetings. The transcript is split by speaker, and you can name a voice so it is recognized in later recordings on the device. The summary has an overview, key points, decisions and action items with owners and due dates when they were said. Action items land in a to-do list. “Fix a word…” corrects a misheard term everywhere in the note and in later recordings.
- What it does not do. No meeting bot, no Zoom, Google Meet or Teams, no phone-call recording, no Android or web app, no team workspace. Transcription and summaries need an internet connection, although recording works without one. The interface is in English. Transcripts are AI output and can be wrong.
- Limits. A recording can be up to one hour. The free plan transcribes the first 10 minutes of each recording, 30 minutes a month; see pricing.
Get Nuvi. The meetings use case shows the flow, and compare sets it next to the bot-based tools.