Extract Action Items from Any Meeting Transcript in 2 Minutes

You finish a 45-minute team meeting. Somewhere in it, your manager assigned three tasks, a deadline was mentioned, and someone volunteered for follow-up research. But where exactly? Manual review means scrubbing the recording, re-listening while taking notes, and still missing the item buried in a side conversation.

The two-minute alternative: a speaker-labeled transcript plus one AI prompt.

AI Prompt: Meeting Action Item Extractor

Copy and paste this into ChatGPT, Claude, or Gemini:

Analyze the following meeting transcript and extract all action items,
decisions, and follow-ups.

For each action item, provide:
1. **Task**: Clear description of what needs to be done
2. **Owner**: Person responsible (from speaker labels, or "Unassigned" if unclear)
3. **Deadline**: Due date if mentioned, or "Not specified"
4. **Context**: Brief note on why this was assigned (1 sentence)

Format the output as a structured list organized by owner. At the end, include:
- **Key Decisions Made**: List any decisions finalized during the meeting
- **Open Questions**: Items that need follow-up but weren't resolved
- **Next Meeting Topics**: Items explicitly deferred to future discussion

TRANSCRIPT:
[Paste your transcript here]

This prompt and 120+ others are maintained in our free AI prompts library, in Markdown and automation-ready YAML.

The workflow

1. Get a transcript with speaker labels. Upload the recording to TranscribeBee — processing takes ~2 minutes per meeting hour at $2 per audio hour, and the TXT output labels each voice (Speaker 1:, Speaker 2:).
2. Paste prompt + transcript into your AI chat of choice.
3. Skim the output, fix names (map "Speaker 2" to "Dana" — or tell the AI who each speaker is and it will do it), and send.

Total time from "meeting ended" to "action items in everyone's inbox": under five minutes.

Why speaker labels are the difference-maker

Without them, the AI sees one undifferentiated wall of text and "I'll handle the vendor follow-up" is unattributable. With labels, every commitment has an owner. If your current meeting tool's transcript doesn't separate speakers, that — more than accuracy — is what to fix first.

Customization tips

• For your boss specifically: add "Format this as a brief email to my manager summarizing the meeting outcomes" and the same extraction arrives pre-packaged.
• For task trackers: add "Output as a Markdown table with columns Task / Owner / Due / Context" (or ask for CSV) for clean import.
• For recurring meetings: append last week's action list and ask the AI to mark items as completed, carried over, or newly added — instant standup continuity.
• For long meetings: if the transcript exceeds the chat's input limit, split it in halves and merge the outputs — the prompt works identically per chunk.
• For soft commitments: add "include tentative commitments like 'I can look into that' flagged as UNCONFIRMED" so half-promises don't evaporate.

FAQ

What if nobody stated deadlines? The prompt outputs "Not specified" rather than inventing dates — a useful signal that your meetings need a deadline habit, not a better prompt.

Does it work without speaker labels? It extracts tasks but can't attribute owners reliably. Labeled transcripts are the difference between a to-do list and an accountability system.

How accurate is extraction? Near-complete on explicit assignments; the review pass exists for implicit ones. Skim against your memory once — still 10x faster than writing minutes by hand.

Can it run automatically? Yes — the prompts repo has a YAML version designed for automation pipelines.