The quality of AI minutes depends far more on the transcript going in than on the model writing them. Prepare in this order and the difference is obvious.
Five minutes before you record
- Pick the mode that matches the purpose — meeting, interview, lecture. It changes how the summary is structured.
- Enter the number of speakers you expect. Speaker separation gets noticeably more accurate.
- Point the microphone at the people talking. A laptop's built-in microphone degrades sharply past three participants.
During the meeting
Interim and confirmed sentences look different. Only confirmed ones are translated and summarised, so you can ignore the faint line moving along the bottom.
When a decision is actually made, have one person say so out loud — "so we are going with A". It is the single habit that improves the summary most.
After the meeting
- 1Stopping generates the summary, decisions, action items and open questions.
- 2Each item links to the transcript position it came from, so a disagreement can be settled against the original in one click.
- 3Assign owners to action items and export to Word, PDF or Markdown to share.
If your team is multilingual
With live translation on, every confirmed sentence is translated as it lands and kept beside the original. The minutes are stored with their translations too, so nobody has to run the document through a translator afterwards.