Request access →

What is an AI note taker, and how does it work?

Four steps sit behind every AI meeting note: capture, transcription, speaker labelling and summarising. Each one can go wrong in its own way.

Guide6 min readUpdated The Shin teamAI note takers
In short
  • Capture: audio from a bot in the call, the platform, or your device.
  • Transcription: speech-to-text turns audio into words.
  • Speaker labelling (diarisation): working out who said what.
  • Summarising: a language model turns the transcript into notes and action items.

1. Capture

The tool needs the audio. It gets it in one of three ways: a bot joins the call, the meeting platform's own AI listens, or an app on your computer or phone captures what you hear and say. See built-in, bot or on-device.

2. Transcription

A speech-recognition model turns audio into text. Quality depends heavily on microphones, background noise, accents, people talking over each other and specialist vocabulary. Names, product terms and numbers are where errors cluster.

3. Who said what

Diarisation splits the transcript by voice. Some tools label "Speaker 1, Speaker 2"; others match voices to names using the calendar invite or the platform's participant list. Wrong labels are one of the most common and most damaging errors, because they put words in the wrong mouth.

4. Summaries and action items

A large language model reads the transcript and writes notes: key points, decisions, next steps. This is the most useful step and the riskiest. Summaries can smooth over disagreement, merge two numbers, or assign an action to the wrong person. Always check against the transcript for anything you'll act on. More in how accurate are AI meeting notes.

AI note taker vs transcription vs meeting assistant

  • Transcription gives you the words.
  • An AI note taker gives you the words plus a summary and actions.
  • An AI meeting assistant goes further: help during the call, follow-up drafts, search across meetings. See the difference.

What still needs a human

Deciding what mattered, checking the numbers, and sending the follow-up. The best use of AI notes is to let you listen properly, then spend five minutes reviewing instead of thirty writing.

Questions people ask

What does an AI note taker do?
It captures meeting audio, transcribes it, labels speakers, and uses a language model to write a summary with decisions and action items.
How do AI note takers know who is speaking?
Through diarisation, which separates voices, sometimes combined with calendar or participant data to attach names. It can make mistakes, especially with similar voices or crosstalk.
Are AI meeting notes accurate?
Usually good at the overall gist and weaker on names, numbers and who owns an action. Review them before sharing.
Do I still need to take my own notes?
A few lines of your own help: they tell you, and the AI, what mattered most.

About Shin. We’re building something for people whose work happens in conversation. Private beta, by invitation.

Request access →

Keep reading

AI note takers · Guide

How accurate are AI meeting notes? Where they go wrong and how to check

AI notes are usually right about the gist. The errors that matter hide in the details.

5 min read →AI note takers · Guide

Bot vs bot-free AI note takers: how they differ

The biggest design choice in AI note-taking isn't the AI. It's how the audio gets in.

6 min read →AI note takers · Guide

AI meeting assistant vs AI note taker vs transcription: what's the difference?

Three labels, often used interchangeably. They describe different jobs.

4 min read →
More AI note takers guides →