How to turn messy meeting notes into action items with ChatGPT (full teardown)

Every team ends a client call the same way: someone has notes, nobody has actions. This is the playbook that fixes it, torn down step by step on a real call, with the real input, the real output, and the two places where a human has to stay in charge.

The input: the actual notes, untouched

Tom’s notes from a 45-minute call between Brightside and their client Fernhill Coffee, typed live, pasted with zero cleanup:

fernhill call 12 june — maya, tom, léa / dana + marco (fernhill)

packaging v2 — dana likes direction B mostly, wants the brown warmer?? priya to try 2 variants
marco: holiday gift boxes AGAIN this year, bigger run. need designs earlier than last yr (last yr was "a disaster" his words)
social — léa showed sept content calendar, approved except the meme-y ones, dana says "we're playful not silly"
dana asked about photoshoot budget — maya said we'd send options
!!! they're launching a decaf line?? first we've heard. marco says details "in a few weeks"
someone needs to send them the brand guidelines pdf again, they lost it
invoice question from dana — she'll email finance thing separately
next call in 2 wks same slot

Note the mess: an action with no owner (“someone needs to send”), a request with no deadline (the photoshoot budget options), and a surprise (the decaf line) that is information, not action.

The run: prompt M1

Paste the notes into prompt M1, placeholders filled. Its central rule, the one that makes the difference: if the owner or deadline was not said in the meeting, the output writes TBC instead of inventing one.

The output: what the AI produced

Summary Fernhill reviewed packaging direction B and approved it with a warmer brown to be explored. Holiday gift boxes are confirmed again this year at a larger run, with designs needed earlier than last year. The September social calendar was approved except the meme-style posts. Fernhill is launching a decaf line, details to come in a few weeks. Photoshoot budget options were requested.

Decisions made

• Packaging direction B approved as the base; explore warmer brown variants.
• September social calendar approved, minus the meme-style posts.
• Holiday gift boxes going ahead, larger run than last year.

Action items

Action	Owner	Deadline
Design 2 packaging variants with warmer brown	Priya	TBC
Send photoshoot budget options to Dana	Maya	TBC
Remove meme-style posts from Sept calendar	Léa	Before Sept calendar ships
Re-send brand guidelines PDF to Fernhill	TBC	TBC
Schedule next call	Tom	In 2 weeks, same slot

Open questions

• What are the specifics and timeline of the decaf line launch?
• What does “earlier than last year” mean concretely for gift box design deadlines?
• Dana’s invoice question: arriving separately by email.

Notice what the AI did not do: no invented owner for the brand guidelines PDF, no invented date for the packaging variants, because nobody said them. That is the prompt working, not failing. The decaf line landed in open questions, not as a made-up action.

The chain: prompt M2 writes the follow-up email

The playbook does not stop at the summary. M1’s output feeds a second prompt that drafts the client follow-up:

Subject: Fernhill: recap & next steps

Hi Dana, Hi Marco,

Thanks for today, good session. Where we landed:

• Packaging: direction B approved; Priya will explore two warmer-brown variants.
• Social: September calendar approved, minus the meme-style posts.
• Gift boxes: confirmed for this year, bigger run, designs earlier than last year; we will propose a timeline.

From you: decaf line details when ready, and Dana’s invoice note to our finance email. From us: photoshoot budget options, and your brand guidelines PDF (re-sending today).

Next call in two weeks, same slot; invite to follow.

Maya

The human checkpoint: 30 seconds that keep it honest

Tom assigned the TBC items in 30 seconds (guidelines PDF → Sam; variant deadline → before next call). Maya softened “minus the meme-style posts” to “we’ll dial the playful posts to fit your voice”, a client-relationship judgment no prompt should make for you. Then it shipped.

That is the principle behind every playbook in the kit: the AI does the structure and the first draft, the human makes the judgment calls, and the prompt is written to make those calls visible instead of burying them under plausible text.

The clock

Before: about 25 minutes (rewrite notes, chase who-owns-what, draft the email). After: about 3 minutes including review. At 4 client calls a week, that is an hour and a half back, every week, for whoever takes the notes.

Run this yourself

Prompt M1 is public and free, like the other 49 in the library. The full P1 playbook (the M1 → M2 chain, the summary template, the checkpoints) is part of Kit IA, along with 9 more playbooks built on the same logic: proposals, weekly reporting, client onboarding, hiring.