Use meeting notes to test a smaller recurring meeting
Everyone agrees the weekly ops meeting runs long, and nobody can say what to change, because the case against it is a feeling. Six sets of its own notes, read side by side, will show whether it is producing decisions or producing attendance.
What you'll make
a source-linked sheet of decisions, actions and open items from one recurring meeting, and a short, reversible proposal to the person who runs it.
What you'll need
six existing sets of notes from one series that you may read (AI summaries or hand-written notes both work), the meeting’s purpose, and ChatGPT, Claude, Gemini or Copilot in your work account. About 40 minutes. You don’t need to start recording anything.
Example result.
Six scheduled · six notes available
Explicit decisions: 1 (Leeds run to Thursdays, Dan)
Assigned actions: 1 (carrier shortlist by 9 Oct, Priya)
Open in all six notes: Depot 3 cover
Proposal: name an owner for Depot 3 cover, then try 30 minutes
for four weeks with the blocker slot kept. Review on 6 November.1. Pick one series and one question
Choose one ordinary operations or project meeting with one owner. Write down its purpose and the single question you want the notes to answer: does it produce decisions, do the decisions need everyone invited, or do decisions stick?
Take consecutive instances rather than the six most frustrating ones. Record cancelled meetings and missing notes as gaps, not as zero-output sessions. Don’t use this for one-to-ones or sensitive conversations, and check that reusing the notes for a process review is allowed. Meetings you should never record covers that line.
Why this works: one question decides what you extract. A sheet answering all three is a document nobody reads, which is an ironic end to a project about meetings.
What you should see: a named series, its purpose in one sentence, one question and a list of six dates.
2. Collect the notes and keep their dates
Open each set of notes and keep its title, date and link. If you can’t open one, ask the organiser rather than working around it. In Google Meet, seeing the notes attached to a Calendar event doesn’t mean you can open them: the document’s own sharing settings decide that. Use the fullest record available: in Meet’s notes document, that’s the Full notes tab rather than Quick notes.
Remove anything personal the review doesn’t need.
Why this works: dates and links let you check the table later, and the fuller record keeps the distinctions you’re about to count.
What you should see: six sources, each with a date and location, and any missing ones marked as missing.
3. Ask for an extraction you can trace
Open a new chat in your work account, paste the labelled notes and use:
Use only these notes from [number] consecutive instances of [series].
Purpose of the meeting: [purpose]. Question for this review: [question].
Make a table: meeting date, explicit decision, agreed action, owner/date, open item, source location. Use one row per item.
Quote decision wording exactly. Keep conditions and uncertainty. Write "none recorded" where the notes contain no item and "source unavailable" for missing notes.
A decision is something the meeting agreed. A task given to a person is an action, not a decision. Do not judge anyone's contribution or the meeting's value, and do not recommend cancelling it.
NOTES:[number] and [series] identify the meeting, for example “six” and “Thornbury weekly ops sync”. [purpose] and [question] are what you wrote in step 1. Paste the notes after NOTES:.
Why this works: quoting stops a discussion being promoted into a decision. “None recorded” gives the model a legitimate way to produce nothing, which is the best defence against it producing something.
What you should see: one row per item, every row with a source location, and “none recorded” where the notes are empty.
4. Read the table for three patterns
- Decision density. How many instances record at least one decision? One out of six is a finding; it isn’t proof the meeting is useless, because coordination and raising problems may be its purpose too.
- Owner concentration. How many distinct names own decisions? If two people decide everything, the meeting may be working for two people. It doesn’t tell you who contributed.
- Open-item recurrence. Does the same item stay open week after week? This is the most useful pattern and the least visible: each individual week, the item felt like progress.
Before counting a recurrence, check it is the same issue. “Carrier quotes” one week and “carrier transition” the next may be different stages.
Why this works: each pattern points to a different change: a written update, a smaller invite or an owner.
What you should see: three counts, each with the rows behind it.
5. Propose one change and a way to reverse it
Take the sheet to the organiser first. A stuck item may need an owner rather than a shorter meeting. If a trial makes sense, propose one change, an end date and a stop condition, such as delayed urgent decisions or someone’s access needs not being met. Track the extra follow-up too: fewer minutes aren’t time saved if people spend longer chasing answers afterwards.
Why this works: a reversible experiment is easy to agree to and doesn’t ask anyone to admit their meeting was a waste.
What you should see: a three- or four-sentence note with a question, one change, a review date and a way back.
Check the result
- Every decision in the table has a quoted source line you have opened yourself.
- Tasks given to one person are counted as actions, not decisions.
- The counts state their denominator: “one decision in six notes” or, if two are missing, “in four available notes of six scheduled”.
- The proposal names a review date and a reason to restore the old format.
The usual failure: an assigned task is counted as a decision, so the sheet reports more decisions than the meeting made, and owners look more spread out than they are. Spot it by asking of each “decision”: did the meeting agree something, or did one person agree to do something?
Correct [row] using [source passage]. Classify it as [action / discussion / explicit decision], keeping the exact owner, date and condition. Replace any claim that no decision happened with "none recorded in these notes" unless the source says more. Recount from the corrected rows and show what each count includes.[row] is the row’s date, [source passage] the quoted line from the notes, and the third bracket is the category you believe is right.
Worked example: six notes reveal an unresolved decision
Fictional example. Thornbury Logistics holds a weekly 60-minute operations meeting to coordinate route changes and surface blockers. These are excerpts from six consecutive notes:
28 Aug: Depot 3 cover remains open. No decision recorded.
04 Sep: Agreed to move the Leeds run to Thursdays from October.
Dan owns the change. Depot 3 cover remains open.
11 Sep: Depot 3 cover and carrier quotes remain open.
18 Sep: No route decision recorded. Depot 3 cover still unresolved.
25 Sep: Priya agreed to prepare the carrier shortlist by 9 October.
Depot 3 cover remains open.
02 Oct: Depot 3 cover remains open. Leeds customer communication
still needs confirmation. No new route decision recorded.The first extraction listed two decisions, the Leeds move and “Priya to own carrier shortlist”. The repair prompt, run on the 25 September row, reclassified it:
Date | Decision | Action / owner | Open
28 Aug | none recorded | none recorded | Depot 3 cover
04 Sep | "move the Leeds run| change: Dan | Depot 3 cover
| to Thursdays" | |
11 Sep | none recorded | none recorded | Depot 3; carrier quotes
18 Sep | none recorded | none recorded | Depot 3 cover
25 Sep | none recorded | shortlist: Priya, | Depot 3 cover
| | by 9 Oct |
02 Oct | none recorded | none recorded | Depot 3; Leeds comms- One decision, not two. Priya agreeing to prepare a shortlist is an action. Counting it as a decision would have made the meeting look twice as decisive.
- Depot 3 cover is open in all six. The notes don’t say why. It may need a decision elsewhere or work already happening outside the meeting, so the proposal asks before diagnosing.
- The proposal at the top of the page asks who owns Depot 3, keeps the blocker slot and has a review date. It doesn’t claim the notes proved waste, and keeping the current format is still a reasonable outcome.
Task kit for this job: Download (ZIP) · Reusable instructions
Go deeper
Optional training: Measure the trial without measuring people
Agree the measures before the trial starts, and keep the meeting, not the attendees, as the thing you measure.
Trial change:
Period and meeting owner:
Purpose to preserve:
Baseline evidence and its gaps:
Measures during trial:
Extra work outside the meeting:
Stop/reversal condition:
Review date and decision:Don’t score attendees by talking time or how often the notes mention them. If the trial runs through an unusually quiet or busy month, say so in the result.
Optional training: Use Google Meet's Decisions section
If your notes come from Google Meet’s Take notes for me, open the Full notes tab. It has a Decisions section that tracks outcomes with statuses: Aligned, Needs Further Discussions, Disagreed or Shelved. Google lists it for all users in English and for Beta users in several other languages.
Where that section exists, your sheet is half-built. A string of “Needs Further Discussions” against the same topic is the open-item pattern without you having to assert it. Still open the notes around it: Google says notes can be incomplete, for example after connectivity problems or in meetings shorter than 15 minutes. A short stand-up with three lines of notes has been proven short, not useless.
Optional training: Meetings this method can't judge
Some meetings look thin on decisions because decisions aren’t the output. One-to-ones, check-ins after a difficult period and sessions where the point is that people hear something at the same time will all score badly here, and the score will be confidently wrong.
Don’t treat absent notes as absent decisions, either. Sensitive discussions move out of recorded meetings for good reasons, and the sheet shows that as silence. If the review starts to become “who talks and who doesn’t”, stop: you’re building a performance record out of notes nobody agreed to be measured by.
Sources
- Google Meet Help: Take notes for me: Quick notes and Full notes tabs, the Decisions section and its statuses, Calendar attachments versus document access, and why notes can be incomplete.
- OpenAI Help: Does ChatGPT tell the truth?: why extracted claims need checking against the source.