TL;DR
- Manually asking Claude to summarize Zoom calls produces good output but inconsistent format, different section headers, different task structure, nothing standard call to call.
- A template with embedded output instructions locks the format: same sections every time, action items tagged with effort levels (@quick, @medium, @deep).
- Saving the file with a TRANSCRIPT- prefix is the only trigger. The automation picks it up, writes the structured note, and appends action items to your task list.
- At ten calls a week, one setup session replaces a manual task you’d otherwise run fifty times a month.
I was running the same task at least three times a day. Copy a Zoom transcript out of Zoom, open a Claude session, paste it, ask for a summary. The output was good. It just wasn’t the same twice.
Sometimes the summary came back with “Action Items.” Sometimes “Next Steps.” Sometimes it gave me effort estimates for each task, sometimes it didn’t bother. Occasionally it skipped the Key Decisions section entirely if Claude decided there weren’t any worth surfacing. After ten meetings in a week, I had ten notes that were all slightly different shapes. Scanning them later meant re-reading, not just checking.
The inconsistency was the problem, not the quality. Good output in a non-standard format is almost as hard to use as bad output.
Step 1: Build the template.
The template lives in my vault at 04 Templates/TRANSCRIPT-TEMPLATE.md. It has two parts.
The first is frontmatter, meeting name, date, attendees. You fill that in before the call ends or right after. Takes about thirty seconds.
The second is a set of instructions embedded directly in the file. These tell Claude exactly what to produce from the transcript:
- Write the summary to a specific file path in the Daily Logs folder, using a fixed structure: Summary, Key Decisions, Action Items, Notes
- Tag every action item with
#work - Add an effort level to every task:
@quick(under 10 minutes),@medium(15–30 minutes),@deep(45+ minutes) - Append all action items to
ACTIVE-TASKS.md
Claude doesn’t decide the format. The template decides the format. That’s the fix. Every call comes out the same way because the instructions are the same every time.
The template is 52 lines. Most of them are those instructions.
Step 2: Fill it in, paste the transcript, save the file.
After a call ends, I open a copy of the template, update the frontmatter, and paste the full Zoom transcript at the bottom under the ## Transcript heading.
The only thing that matters about the filename is the prefix. Any file saved with TRANSCRIPT- at the start gets picked up by the Queue Processor, an n8n workflow running continuously on a Mac Mini I keep as an always-on home server. Drop the file in the vault, and the automation handles the rest.
If you don’t have an always-on server running n8n, the template still works. Paste it directly into a Claude session with the transcript included and you’ll get the same consistent output. You just run it manually instead of dropping a file.
Step 3: The Queue Processor writes the note.
The n8n workflow reads the file, follows the embedded instructions, and writes the structured meeting note to the Daily Logs folder. Action items get appended to ACTIVE-TASKS.md with their tags and effort levels intact. A confirmation file lands in Generated/ so I know it ran.
The whole thing takes less than a minute. I’m usually in the next meeting by the time the confirmation comes through.
Outcome:
The workflow has been running three days. Ten transcripts have come through without me touching a single one.
Every note has the same structure: Summary at the top, Key Decisions below it, Action Items tagged and sorted by effort, Notes at the bottom. Every action item is ready to filter in the task list, no cleanup, no reformatting, no deciding whether this one gets #work or not.
One thing that surprised me: the effort tagging turned out to be more useful than the summary. I already remember roughly what was discussed. What I need to know is which tasks are going to take an hour and which ones take five minutes. The template forces Claude to make that call on every item. It’s been right more often than I expected.
The part I was doing manually (the copy, the paste, the prompt, the inconsistent output) is gone.