I used to think I was bad at taking meeting notes.
Then I looked at my notes from a real Tuesday meeting and realized the problem wasn't my note-taking. It was that my notes weren't notes at all. They were a transcript of my own confusion.
Here's an actual sample from my notebook, copied word for word:
"budget thing - Dana? - Q3 maybe - check w/ R - shipment issue again - someone said Friday?? - follow up - Mike mentioned something about the template - unclear - ASK ABOUT THE THING"
That's it. That's the whole entry. Forty-seven minutes of a meeting, and what I had to show for it was a phone number with no name, two question marks, and the phrase "the thing."
If you've ever walked out of a meeting feeling like you understood everything and then opened your notes an hour later to find cryptic hieroglyphics, this post is for you.
Because here's what I figured out: the gap isn't between good note-takers and bad ones. The gap is between notes that capture what was said and notes that capture what was decided. And ChatGPT is genuinely good at closing that gap—as long as you don't let it start inventing things.
Let me show you exactly how I do it.
The Real Problem With Meeting Notes
Most of us take notes like we're recording a podcast. We write down what people said, in the order they said it, and hope it makes sense later.
But nobody ever needs a record of what was said. They need to know three things:
What did we decide?
Who is doing what?
By when?
That's it. Everything else is background noise. And the reason my notes were useless wasn't that I wrote them badly—it was that I never converted them from "what happened" into "what now."
That conversion step is the part everyone skips, because it's tedious and takes forever and nobody wants to do it at 4:30 on a Tuesday.
So I stopped doing it manually.
My Exact Prompt (Steal This)
Here's the prompt I've been using for about eight months now. I keep it in a note on my phone called "Things I Never Want to Figure Out Twice," which is exactly what it sounds like.
"Below are my raw meeting notes. Turn them into four sections:
Decisions Made — only things that were actually decided, not discussed.
Action Items — each with an owner and a deadline.
Open Questions — anything unresolved or unclear.
Parking Lot — topics mentioned but not addressed.
Rules: Do not invent any names, dates, or deadlines. If an owner or deadline isn't clear from my notes, write 'UNASSIGNED' or 'NO DATE' instead of guessing. If something is ambiguous, put it under Open Questions rather than deciding for me."
That last paragraph is the entire ballgame. Let me explain why.
Why "Do Not Invent" Is the Most Important Line
ChatGPT wants to be helpful. That's the whole design. So when your notes are vague, it fills in the blanks—and it fills them in confidently, in clean, professional language.
That's the worst possible outcome. A wrong guess written badly is easy to catch. A wrong guess written well slides right past you and into your manager's inbox.
I learned this the hard way. About six months ago, I pasted in notes from a project sync and asked for a summary. ChatGPT handed back a clean bullet list that included a deadline I had never written down. It wasn't in my notes. It wasn't in the meeting. It just... appeared, because the sentence structure implied a timeline and the model helpfully supplied one.
I caught it because I knew the project. If I hadn't, I would have sent that summary out and looked like I had no idea what was going on.
So now every prompt I write includes an explicit "do not invent" instruction, plus instructions on what to do instead of guessing. "UNASSIGNED" is a useful answer. A fabricated name is not.
Why Four Sections and Not Three
I used to ask for three: decisions, actions, questions. Then I noticed something getting lost—the stuff people mention and then never address.
You know the pattern. Someone says, "We should probably revisit the vendor situation at some point," everyone nods, and it vanishes into the void. Six weeks later it resurfaces as a crisis.
The Parking Lot section catches those. It's not an action item because nobody owns it. It's not an open question because nobody's asking it yet. It's just a thing that exists, and now it's written down where I can see it.
A Real Example, Start to Finish
Let me walk you through an actual run from last month.
The Raw Input
This is what I pasted in. I've changed the names, but the mess is unedited:
"kickoff - new template rollout
Dana opened - timeline concerns
R said current process fine, doesn't want to retrain team
discussed Q3 vs Q4 start
someone mentioned the compliance angle - Janine?
no decision on timing, revisit next week
Mike will draft the comms
need to check with IT about access permissions
follow up on the training doc - who owns it?
parking lot: possible vendor change
Dana wants a status update by EOW
unclear if R is on board"
That's roughly 90 words of fragment soup. Notice how many question marks and vague references are in there.
What ChatGPT Gave Back
Decisions Made
No final decision on Q3 vs Q4 rollout timing; revisit next week.
Current process acknowledged as working; retraining concerns raised.
Action Items
Draft communications — Owner: Mike — Deadline: NO DATE
Check with IT regarding access permissions — Owner: UNASSIGNED — Deadline: NO DATE
Provide status update — Owner: [me] — Deadline: end of week (EOW)
Identify owner for training document — Owner: UNASSIGNED — Deadline: NO DATE
Open Questions
Is R on board with the rollout?
Who raised the compliance angle (notes say "Janine?") — confirm.
Who owns the training documentation?
Parking Lot
Possible vendor change (mentioned, not addressed).
What I Had to Fix
It misread two things. First, it put "provide status update" on me, which was correct—Dana did ask me for that, I just hadn't written my own name down. Second, it treated "current process fine" as a decision when it was really just R's opinion.
Both fixes took about thirty seconds. Total time from paste to finished action list: maybe twelve minutes.
Before ChatGPT, cleaning up those same notes took me closer to thirty-five minutes, and I usually missed something.
The Workflow, Step by Step
Here's the whole thing, in order.
Step One: Take Notes However You Want
I'm not going to tell you to change how you take notes during the meeting. Write fragments. Use initials. Draw arrows. Do whatever your brain does under pressure.
The cleanup is where the structure comes from, not the capture. Trying to take perfectly structured notes in real time is how you end up missing half the meeting because you're formatting bullets.
Step Two: Strip Anything Sensitive
This step is not optional and it's not a formality.
Before I paste anything into ChatGPT, I remove:
Client names and account numbers
Coworker full names, especially in any critical context
Salary, contract, or financial specifics
Anything marked confidential or internal-only
I replace names with placeholders—"Person A," "the client," "the manager"—and I keep the actual mapping in my head. If the notes can't be anonymized without becoming useless, I don't paste them. I do it the slow way.
Step Three: Paste and Run the Prompt
Drop the notes in, paste the prompt, hit enter. Takes about ninety seconds.
Step Four: Review Line by Line
This is the part people skip, and it's the part that matters most.
I read every single line and ask: Is this actually in my notes, or did ChatGPT helpfully add it? Anything I can't trace back to something I wrote gets deleted or moved to Open Questions.
Twelve minutes total. Ninety seconds of generation, the rest spent checking.
Step Five: Send It
I drop the four sections into the follow-up email and send. The whole thing takes less time than my old cleanup process, and it's more accurate.
What This Doesn't Do
Let me be clear about the limits, because this is where people get burned.
It doesn't know what happened in the room. If your notes don't capture a decision, ChatGPT can't recover it. Garbage in, clean-looking garbage out.
It doesn't know your team. It can't tell you that "R" is actually resistant to the whole project, or that Mike always volunteers for comms and never delivers. That context is yours.
It doesn't know what matters. It treats every line in your notes as equally important. You're the one who knows that one action item is a formality and another is a career-defining deadline.
It won't catch its own mistakes. If it invents a deadline, it will present that deadline with the same confidence as everything else. The review step isn't optional. It's the entire safety mechanism.
The Test Card
Task: Convert raw meeting notes into decisions, action items, open questions, and parking lot items.
Prompt or workflow: The four-section prompt above, with explicit "do not invent" instructions.
Starting conditions: Roughly 90–150 words of unstructured fragments, abbreviations, and unclear references.
Time before ChatGPT: 30–40 minutes of manual cleanup, with a meaningful chance of missing something.
Time after ChatGPT: 10–15 minutes total, including full line-by-line review.
Editing required: Moderate. Expect to fix one or two misread abbreviations, and to move at least one item out of "Decisions" that was really just an opinion.
What went wrong: It converted a stated opinion into a decision, and it assigned an owner based on sentence structure rather than actual content. Both were caught in review.
Privacy notes: All names replaced with placeholders. Client identifiers, financial details, and confidential project names removed before pasting. If the notes can't be anonymized, don't use this method.
Who should not use this method: Anyone in a workplace that prohibits pasting internal content into external AI tools. Anyone whose notes can't be meaningfully anonymized. Anyone who doesn't have time to do the review step—because skipping it is worse than not using ChatGPT at all.
The Thing I Keep Coming Back To
The reason this workflow works isn't that ChatGPT is smart. It's that the work of converting notes into action items is mechanical, and mechanical work is exactly what ChatGPT is good at.
The judgment—what was actually decided, who really owns what, which items matter—that stays with me. Where it belongs.
I tried it at my desk so you don't have to. No magic. Just a workflow that survived Monday.
