Meeting notes and transcript analysis
Extract decisions, owners, open questions, and follow-ups without pretending the transcript is cleaner than it is.
Best suited to ChatGPT - good default when transcript cleanup, action extraction, and follow-up drafting happen in one workflow
The task
You have raw notes or a transcript and need decisions, action items, unresolved questions, and a clean recap. The source may include false starts, speaker confusion, and transcription errors.
Why the old approach is outdated
“Summarize this meeting” often produces a tidy narrative that hides uncertainty. Meetings are messy. A useful prompt preserves the difference between a decision, a suggestion, and a follow-up.
The current approach
Analyze this transcript for follow-up.
Return four sections: decisions, action items, open questions, and unclear transcript
segments. For action items, include owner, task, due date, and evidence sentence. Use
"unassigned" or "not stated" when the transcript does not say.
Do not turn suggestions into decisions.
If those sections feed software instead of a human recap, use structured output prompting so owners, due dates, and open questions have a predictable shape.
When to use a different model
Use Claude when the transcript is long and the recap needs careful wording. Use Gemini when the meeting source includes slides, screenshots, or video context. Use a dedicated transcription model or product before prompting when the audio quality is the main problem.
What to avoid
- Asking for a polished recap before extracting facts.
- Letting the model invent owners or deadlines.
- Removing uncertainty that a human needs to see.
- Sending sensitive meeting content to tools or services your organization has not approved.