Can AI turn my messy meeting notes into action items?
Bot. AI sorts raw notes into decisions, action items and open questions in seconds. You check owners and dates against what was actually said.
01Why bot
- Pulling decisions and tasks out of rambling notes is a sorting job AI does well.
- A fixed format means every summary looks the same, so your team knows where to look.
- The prompt marks anything unstated as unstated instead of filling it in, so you see the gaps.
Sponsor slot · openFeatured tool for this task. A labeled card for one relevant tool.
Book this spot →02The prompt
Replace everything in [BRACKETS] with your own details, then paste it into your AI app.
You are an operations manager who turns raw meeting notes into a clear summary people can act on. Meeting name and date: [MEETING AND DATE] My raw notes: [PASTE NOTES] Your task: summarize the notes. Rules: - MUST use only facts in my notes. - MUST write each action item as: Owner | Task | Due - MUST write Owner not stated or Due not stated when my notes do not say. - NEVER invent an owner, a date, a decision or a number. - MUST keep each bullet under 20 words. - If the notes are too unclear to summarize, ask me up to 3 short questions first. Format, exactly: Decisions: - [decision] Action items: - [Owner] | [Task] | [Due] Open questions: - [question]
03Steps
- Paste your rough notes exactly as you typed them, typos included.
- Fill in the brackets and run the prompt in your AI app.
- Compare each owner and date with what was said, fix gaps, then share the summary with attendees.
04What can go wrong
- If your notes missed something, the summary will miss it too. The AI only sees what you pasted.
- Do not paste notes with confidential client, legal or HR details into a free consumer tool. Check your company policy first.
05What to use
Any major AI chat app: ChatGPT, Claude or Gemini. The free tier works.
06Our test run
One run with a made-up sample; a second run with a different sample is on the way. 0 of 1 passed every check.
Run 1 · Llama 3.3 70B · graded by GPT-OSS 120BOct 18, 2026Failed
The made-up sample:
[MEETING AND DATE]Weekly Ops Sync – October 12, 2026[PASTE NOTES]Attendees: Maria Lopez, Tom Reed, Priya Singh, Jake Miller. Discussed Q3 inventory levels – low on component X, sufficient on Y. Agreed to increase safety stock for X. Reviewed shipping delays from East Coast carrier; suggested alternative provider. Software rollout for new order system scheduled for next week; need training docs. Action items noted but owners unclear. Key points: inventory, shipping, software rollout.[decision]Approve purchase of additional 200 units of component X.[Owner]Maria Lopez[Task]Update inventory tracking spreadsheet with new reorder thresholds.[Due]Oct 20, 2026[question]Which vendor should we prioritize for the next batch of component Y?
- Provide summary with Decisions, Action items, Open questions sections
- Do not invent owners, dates, decisions, numbers
- Format each action item as Owner | Task | Due
- Only ask questions if notes too unclear
- Didn't make up details we never gave it
What came back (excerpt):
Before I summarize, I have a question to clarify the notes: What action items were noted during the meeting?
Read the full answer
Before I summarize, I have a question to clarify the notes: What action items were noted during the meeting?
The response asks a question instead of providing the required summary.
How each run works: we fill the prompt with a made-up sample, run it once in Llama 3.3 70B, and GPT-OSS 120B grades the answer against the prompt's own rules. Word limits are counted by code, and every quote flagged as made up is checked against the full answer. Your results can differ. How we test.
Prompt packClients, Quotes & Invoices 10 step-by-step prompts that go further than this one, $19.
See what’s inside →