An EA posted this week asking a question that pulled 40 replies in a day: how do you take notes in a meeting that runs over an hour, when people jump between decisions, dates, and follow-ups, and you're supposed to be present in the room the whole time? She'd tried typing live, keeping a running doc, cleaning it up after. Every method cost her something. Either she missed a moment while writing, or she lost an hour afterward rebuilding what was said.
Here's the reframe this issue runs on, and it's the whole reason the note-taking question is the wrong question: the transcript was never the deliverable. A meeting is a loop, and the loop has four stages. Each stage produces exactly one artifact. Miss a stage and the work leaks out of the gap. Get all four and the meeting turns into results instead of notes, with you as the person who runs the loop, not the person who transcribes it.
The capture is the easy 20%. The loop is the 80% almost nobody runs on purpose. Below is the whole thing, one stage at a time, with the one artifact each stage owes you and the exact way the work disappears when you skip it.
One honest note up front, same as always: a loop will not make people do the things they agreed to do, and it will not fix a meeting that should have been an email. What it does is make the skipping visible. When every action has one name and one date, “I didn't know that was mine” stops working. You stay the judgment. The loop just stops letting things fall through.
Subscribe to The Dispatch and the rest of this issue unlocks right here, along with the full archive. One email every Tuesday, about 8 minutes, built for admin professionals. No spam, unsubscribe anytime.
Already subscribed on this device? This unlocks automatically.
Four stages, one artifact each, and the fourth feeds the first, which is what makes it a loop and not a checklist you abandon by Thursday.
Prep produces the Agenda.
Capture produces the Decisions, Actions, Owners table.
Follow Up produces the Recap.
Close produces the Status Check — and the Status Check is what you build the next Agenda from.
Notice what the recorder is doing in all of this: nothing structural. Any AI note-taker can hand you a transcript. The transcript is raw material for stage 2, not the point. The point is the four artifacts and the discipline of producing the same one every time. AI's job here is to assemble and draft each artifact fast, so the discipline costs you minutes, not willpower. You own the judgment inside each: which item is truly stuck, whose name goes on it, what's actually decided.
You don't need new software or a budget to start. You need one transcript and the four prompts below.
Stage 1 — Prep → the Agenda. The agenda is assembled, not written. Three inputs, in this order: the actions still open, the decisions still pending, and the new items raised since last time. It goes out on a fixed schedule, from the same place, every time. Let AI assemble it from what you already track:
Build the agenda for my next meeting by assembling three sections in this exact order: (1) OPEN ACTIONS — items from my running log still marked Moving or Stuck; (2) PENDING DECISIONS — decisions raised but not yet made; (3) NEW ITEMS — anything added since the last meeting.
Here is my running action log: [paste]. Here are new items raised since last time: [paste]. Do not invent items. Keep it to one page, ordered by what needs a decision first.
Stage 2 — Capture → Decisions, Actions, Owners. Three columns, filled live. One name, one date, never a team. This is where the transcript finally earns its keep, as the raw feed you turn into the table:
Here is the meeting transcript: [paste]. Turn it into a three-column table: DECISION or ACTION | OWNER | DUE DATE.
Rules: every action gets exactly one owner (a person, never a team or "we") and one due date. If the owner or date wasn't stated, put CONFIRM so I can chase it — do not guess. Ignore discussion that wasn't a decision or an action. Do not invent rows.
Stage 3 — Follow Up → the Recap. One message, one channel, a fixed window after the meeting, and the same three headings every single time: what we decided, who owes what by when, what stayed open. Predictable beats polished:
Turn this Decisions/Actions/Owners table into a recap message with exactly three headings, in this order: "What we decided," "Who owes what by when," "What stayed open." Keep it skimmable and under 150 words. Match my tone: [paste 2–3 sentences of how you normally write]. Here is the table: [paste from stage 2].
Stage 4 — Close → the Status Check. Every action gets one of three labels: Done, Moving, or Stuck. Moving items are carried to the next agenda without discussion. Only stuck items earn agenda time. Then those carried and stuck items flow straight back into Stage 1, and the loop closes:
Here is my running action log: [paste]. Here is the latest recap: [paste from stage 3]. Update the log: mark each action Done, Moving, or Stuck based on the recap.
Then output two things: (1) the CARRIED list — everything still Moving, to roll to the next agenda as-is; (2) the STUCK list — everything blocked, which is the only thing that needs meeting time. Keep each item's original owner and due date unless a new one was set.
Notice what none of these four stages ask of you: not to type faster, and not to hand your judgment to a machine. They move the mechanical part — assemble, transcribe, format, chase — off your plate, so the part only you can do gets your full attention: knowing what actually matters, whose name belongs on it, and which of the “stuck” items is the one to spend the room's time on. That's the whole game. Run the loop, don't transcribe the meeting.
1. The tools are good at the wrong half. Every 2026 note-taker roundup landed on the same quiet verdict: nearly all of them nail the transcript, and almost none are good at what happens after it. Which is oddly great news for you, because the transcript was never the valuable part. Stages 1, 3, and 4 are, and those are judgment, not typing. AI just clears the busywork around them.
2. Someone in your shoes already built a loop. An EA posted her own ChatGPT workflow this week (hers was for tailoring job applications) and reported it landed her two interviews in under a month. The takeaway isn't the cover letters. It's that the people pulling ahead aren't waiting for IT or a training budget. They build a small, repeatable system around real work. The meeting loop is the same move, pointed at the thing that eats your Tuesdays.
3. Microsoft quietly fixed the “who owns this meeting” trap. An EA this week described organizing every meeting from her own calendar just so she could edit it, then getting caught out on medical leave when meetings she “owned” couldn't be moved. The new shared and delegate mailbox scheduling means meetings can live on the exec's mailbox identity instead of yours, so they survive your day off. Ask your admin if it's on. One honest limit: Copilot can summarize and draft inside a delegate inbox, but it won't reliably book or move calendar items there yet, so schedule from the delegate mailbox yourself. The free Digital Zen resource library has a starter Meeting Loop pack that pairs with the four prompts above: digitalzen.site/resources
Take a single transcript, paste it into the Capture prompt, and let it hand you a Decisions/Actions/Owners table: one name, one date, gaps flagged CONFIRM. It's the fastest way to feel the difference between having notes and having a loop.
Then hit reply and tell me: which of the four stages does your current process skip? I'd bet Stage 1 or Stage 4. Almost everyone captures and recaps, and almost nobody assembles the agenda or closes the loop. I read every reply.
Master your tools. Own your work.
— Kristy
Digital Zena subsidiary of Fusion AI Consulting LLC
The AI Prompt Library for Executive Assistants, the Meeting Prep and Follow-Up Template, the New Tool Adoption Checklist and the rest of the Digital Zen resources are free and open. No form, no cost.
Open the library →