Meeting notes summarizer prompt builder
Raw meeting transcripts are nearly useless — they are long, full of crosstalk, and bury the three things people actually need: what was decided, who is doing what by when, and what is still open. This builder writes a prompt that turns that mess into a structured summary, with action items captured in an owner-and-deadline format so they are trackable instead of forgotten.
How it works
You choose the meeting type and tick the sections you want — decisions, action items, open questions, key discussion points, and an optional one-line TL;DR. The generated prompt instructs the model to read the transcript, discard filler and false starts, and slot the substance into your chosen sections. The action-item format setting controls whether each item records an owner and a due date, and the prompt tells the model to flag anything ambiguous rather than inventing an assignment.
Why owner + deadline is the critical action-item format
Most meeting notes fail at the point they matter most: they capture what was discussed but not who is committed to doing what by when. The difference between “we need to update the pricing page” and “Sarah to update the pricing page by Friday” is the difference between something that gets done and something that drifts into the next meeting’s backlog.
The prompt this tool generates specifically instructs the model to extract each commitment in owner-action-deadline form. When a transcript contains an ambiguous commitment (“someone should follow up with the vendor”), the prompt tells the model to flag it explicitly rather than guessing or omitting it. Ambiguous ownership is one of the most common reasons action items fall through — surfacing it in the summary makes the ambiguity visible and gives the meeting organiser the information needed to resolve it.
Handling different transcript quality levels
Auto-generated transcripts from Zoom, Teams, or Google Meet vary widely in quality. Common issues and how the generated prompt handles them:
- Speaker misidentification. Transcription software often assigns speech to the wrong participant, especially when speakers have similar voices or talk over each other. Providing the participant list lets the model apply heuristics, but review attributions manually for important action items.
- Technical jargon. Domain-specific terms, product names, and acronyms are frequently mangled in auto-transcripts. The prompt tells the model to infer from context rather than transcribe literally, but specialised content benefits from a post-generation human review.
- Crosstalk and interruptions. The prompt instructs the model to identify completed thoughts and discard false starts, but dense overlapping speech sometimes loses meaning that only a human reviewer can recover.
- Very long meetings. Transcripts from two-hour working sessions can exceed the context window of some models. For long meetings, split the transcript into thematic segments and run the prompt on each, then combine the outputs.
What the TL;DR section does
A one-line summary at the top of any meeting notes document is disproportionately valuable: it is what gets read in the email preview, the Slack message, and the casual scan of a document before someone opens it. The TL;DR should capture the single most important outcome or decision from the meeting in plain language. The prompt generates this first, so even if a recipient reads nothing else, they know whether the meeting produced something they need to act on.
Tips and notes
- Paste the participant list. Giving the model the attendee names helps it attribute action items to real owners instead of “someone.”
- Keep the TL;DR on. A one-line summary at the top is what people actually read; the structured detail below is for whoever needs to act.
- Review owner attributions. Auto-transcripts mangle names — always sanity check who got assigned what before circulating the summary.
- Use it for standups too. Pick the standup type and the model produces a tight per-person update with blockers surfaced separately.