Preparing a meeting with AI: the full method

AI saves time on a meeting at three moments: before (framing the agenda and the questions), during (taking notes for you) and after (writing the minutes and the action list). It replaces neither the facilitation nor the decisions — it removes the preparation and clean-up work, which delegates very well. Here is the method, phase by phase.

Before the meeting: frame it in five minutes

Start by getting the agenda and key questions written. This is the task where AI is most useful and least risky: you give it the context, it structures. Describe the goal of the meeting, the participants and the time available in one or two sentences, then ask for a timed agenda.

Prepare the agenda for a 45-minute meeting with the marketing
team. Goal: decide the Q4 budget. Participants: me, the marketing
lead, the management controller. Propose 4 to 5 timed points,
and for each the decision expected at the end.

Follow up with the questions to ask: ask the assistant for the three or four questions that will move each point forward. You arrive at the meeting with a framework, not a blank page. If you often prepare the same type of meeting (weekly committee, project update), turn this framing into a reusable prompt: you no longer start from scratch each time.

During the meeting: let AI take the notes

Delegate the note-taking to stay focused on the discussion. This is the second concrete gain: instead of typing while others speak, you let the assistant transcribe and you genuinely take part.

The integrated suites do it natively: according to Microsoft’s documentation, Copilot in Teams produces a summary and a record of the meeting; on the Google side, Gemini in Meet offers automatic note-taking. For a meeting outside these environments (in person, on another tool), dedicated transcription tools take over.

A non-negotiable reflex: warn the participants. An automatic transcription is a recording — announce it at the opening, exactly as you would for a normal recording. It is a matter of trust, and sometimes of law.

After the meeting: minutes and actions

Turn the raw transcript into usable minutes in one request. This is where the clean-up, long the organiser’s chore, almost entirely disappears. From the notes or the transcript, ask for three distinct deliverables:

From these meeting notes, produce:
1. a 5-line summary for those who were not there;
2. the record of decisions made (one line each);
3. the action list: who does what, by when.
Flag any open item without deciding in my place.

The instruction “flag the open items without deciding” is important: it stops the assistant from filling the gaps by inventing a decision. Once the minutes are reread, you can have them formatted and sent by email to the group.

The method at a glance

Phase	What AI does	What stays yours
Before	Timed agenda, key questions, brief	Set the goal, choose the priorities
During	Automatic transcription and notes	Facilitate, arbitrate, decide
After	Summary, record of decisions, action list	Reread, correct, distribute

The limits to keep in mind

The generated minutes are a first draft, not a source of truth. An assistant can misattribute a sentence, confuse a hypothesis with a decision, or misrepresent a technical exchange. Before distributing, at least reread the record of decisions and the action list — these are the parts that commit people. The checklist to verify an AI answer applies as is to minutes.

Second point: confidentiality. A meeting often contains sensitive information (figures, HR, strategy). Before entrusting a transcript to a tool, check how it handles your data — our confidentiality habits for AI at work detail what to look at.

Key takeaway

AI does not run the meeting for you: it removes the work around it. Frame the agenda and the questions before, delegate the note-taking during (while warning the participants), and have the summary, decisions and actions produced after — then reread what commits people. The time you no longer spend preparing and cleaning up, you spend listening and deciding.