When to use it

The signal is precise: in today’s meeting someone said “wait, hadn’t we already decided this?”. If that happens, last week’s decisions evaporated somewhere in the notes, and this playbook is for you. It works with any raw material: your own notes, an automatic transcript, even the side chat of the call.

What you need

Any AI assistant (Claude, ChatGPT, Gemini: the method is identical) and the meeting material, as is. If you want a dry run before using it on a real meeting, the course provides a sample transcript with the expected result to check yourself against (both in Italian, which makes them a realistic multilingual exercise too).

The steps

1 · Gather the raw material. Notes, transcript, chat: everything together, uncleaned. Cleaning is the assistant’s job, not yours. Every minute spent formatting by hand is a minute taken away from thinking.

2 · Ask for decisions, not a summary. The summary is the wrong output: it is the meeting retold in fewer words. What you need is separation into three categories, with ambiguity declared:

From the notes below extract three separate lists: 1) decisions made (only the ones actually closed, with the final outcome if several proposals were discussed), 2) open or postponed questions, with who picks them up, 3) actions with owner and deadline, including constraints mentioned in passing (“after the trade fair”, “when X is back”). If a point is ambiguous, put it in a fourth list called “to clarify” instead of guessing.

3 · Resolve the ambiguities: you, not the AI. The “to clarify” list is the most valuable part of the output: that is where your judgment is needed. Resolve them before distributing anything, by asking the people who were there. An ambiguity hidden today is an expensive misunderstanding two weeks from now.

4 · Turn actions into trackable commitments. Every action goes where the team actually works: board, issue tracker, calendar, channel message. Ask the assistant for the exact format of that tool (“write these actions as Trello cards, title plus description” or “as a Slack message for the project channel”), so you never rewrite anything by hand.

5 · Close the loop at the next meeting. Open the next meeting from the list of commitments, not from a blank page. Two minutes: done, not done, why. This is where the workflow becomes a habit and meetings stop repeating themselves.

A complete example

On the sample transcript (25 minutes, four people, a service launch) step 2 must produce: four decisions, including a price that changes from 280 to 350 euros during the discussion; six tasks, one of which carries a constraint mentioned in passing (the copy review only after the 20th, because the owner is at a trade fair before then); and one question explicitly postponed, the legacy clients to realign with the new price list. The comparison document lists everything, with the three points where automatic summaries fall down. Run the loop and compare: if your step 2 caught all three, you are ready for real meetings.

Typical mistakes

Trusting the first proposal. When a number gets negotiated in a meeting (280, then 350), the assistant tends to report the one discussed longest, not the one approved. The defense lives in the step 2 prompt: “with the final outcome if several proposals were discussed”.

Promoting a postponed question to a decision. “Let’s discuss it next time” is not an outcome. If it ends up among the decisions, the summary has decided on the meeting’s behalf.

Distributing without step 3. The assistant’s output looks finished, and that is its most dangerous quality: the “to clarify” list must be resolved before the document circulates, not after.

The team variant

Once the playbook works for you, the natural next step is that the summary reaches every participant within an hour of the meeting, in the same format every time. At that point the step 2 prompt becomes a shared saved instruction (a Claude Project or a shared team GPT) and the note-taker can rotate week by week without the result changing. The skill “Minutes → decisions” packages this playbook, and the chapter “Instructions that work” explains how to write step 2 well.