Meeting Transcript → Action Items
The Setup
Every meeting generates the same promise: “I’ll send notes and next steps.” And every meeting is followed by the same reality: the notes don’t get written because the next meeting started and the day kept going.
I was losing follow-ups. Not because I forgot what was discussed — I usually remembered — but because remembering isn’t the same as documenting. A week later, “I think we agreed to…” is not the same as a clear list of who owes what by when.
The Build
The system takes a meeting recording, transcribes it, runs the transcript through AI to extract a summary and action items, and delivers the output to Notion or email. The whole thing runs automatically after every meeting.
The transcription layer handles the raw audio — tools like Otter or the built-in transcription in Zoom/Google Meet. The AI processing layer (ChatGPT or Claude, depending on the workflow) takes the transcript and produces three things: a 3-sentence summary of what was decided, a list of action items with owners and deadlines, and any open questions that need follow-up.
The prompt matters more than the pipeline. “Summarize this meeting” gives you garbage. “List every commitment someone made, who made it, and when they said it would be done” gives you something actionable.
The Mess
The early versions confused discussion with decision. If someone said “we could try running a social campaign next month,” the system would log “Run social campaign next month — Owner: [whoever said it].” But that wasn’t a commitment. It was thinking out loud.
Tuning the prompt to distinguish between “someone explored an idea” and “someone committed to an action” took real iteration. The fix was adding a filter: only log something as an action item if the transcript contains a clear commitment — “I’ll do X by Y” or “let’s make sure Z happens before the meeting.”
The Result
I send meeting notes within an hour of every call. They’re clear, they’re accurate, and they hold people (including me) accountable. Clients have commented on it. “You’re the only person who actually sends action items after a call.”
The time savings are real but the trust savings are bigger. When you consistently follow through on the small things — like writing up what you agreed to — people trust you with the big things.
The Takeaway
The transcript is useless. The action items are the only thing that matters. Build the system that extracts what people committed to, not what they talked about.
How It’s Built
- Transcription: Zoom/Google Meet built-in, or Otter
- Processing: ChatGPT or Claude (custom prompt)
- Delivery: Notion page or email
- Automation: Zapier or manual trigger
- Output: 3-sentence summary + action items with owners + open questions
- Cost: Minimal (transcription tools + AI API costs)