In one minute
- What you will be able to do: turn rough meeting notes into a list of actions, each with an owner and a date.
- What you need: your own notes (or a transcript you are allowed to use) and a free account with an AI chatbot.
- How long it takes: about 10 minutes the first time, 3 to 5 minutes after that.
- The golden rule: the AI drafts, you check. Nothing goes out until a person has read it.
Why it is worth doing
Most meetings end with good intentions and vague notes. A week later nobody is sure who said they would chase the quote, or whether “next week” meant Monday or Friday. Writing up actions is dull, so it often does not happen.
An AI chatbot is good at the dull part: pulling scattered promises and decisions out of messy text and putting them in a neat table. It is not good at knowing what people meant, and it can fill gaps with guesses. So the method below makes the AI show where each action came from, and leaves the final check to you.
Which tool to use
Any general chatbot will do. ChatGPT, Claude, Google Gemini and Microsoft Copilot all have free versions. If your workplace or school has an approved AI tool, use that one: it may come with stronger privacy terms, and putting work material into an unapproved tool can break your organisation’s rules.
Some video-call apps can take notes for you. According to Google’s help page, as of 19 Nov 2026, Google Meet’s “Take notes for me” needs “an eligible Google Workspace edition or Google AI plan”, is only available on accounts set up through work or school, and shows a pencil icon on every participant’s screen while it runs. Zoom and Microsoft Teams offer similar features on some plans; check their own help pages for what yours includes. This guide covers the method that works everywhere: giving your own notes to a chatbot.
The steps
- Check you may use the notes. For a work meeting, check your organisation’s AI policy first. If you want to use a recording or transcript, tell everyone in the meeting before you record. Rules on recording differ between countries and workplaces.
- Strip out what the AI does not need. Delete phone numbers, addresses, pay, health details, customer names and anything confidential. You can swap full names for first names or roles (“Sam”, “finance lead”).
- Decide on the training setting. What you paste is sent to the company that runs the chatbot. Several makers let you stop your chats being used to improve their models: in ChatGPT, OpenAI’s Data Controls FAQ says to go to Settings > Data Controls and turn off “Improve the model for everyone”; in Claude, Anthropic’s Privacy Center points to the “Help improve our AI models” toggle under Settings > Privacy; for Gemini, Google’s Gemini Apps Privacy Hub explains the Keep Activity setting. These settings do not stop your notes reaching the company; they change what it may do with them.
- Paste your notes with a clear request. Use the prompt below. Asking for “Not agreed” instead of a guess is the most important line.
- Make it show its working. The “Source line” column means every action points back to the words in your notes it came from.
- Check every row against your notes. Fix names and dates, delete anything you cannot find in the notes, and turn every “Not agreed” into a question for the group.
- Send it yourself. Copy the checked list into an email or shared document and ask people to flag mistakes within a day.
A prompt you can copy:
Below are my notes from a meeting on [date]. Make a table of actions with four columns: Action, Owner, Due date, Source line. Only include actions that are clearly in the notes. If an owner or date is missing, write “Not agreed” rather than guessing. For Source line, quote the words from my notes. After the table, list any decisions made and any open questions. Notes: [paste your notes here]
A worked example
Here are some typical rough notes from a volunteer group planning a charity quiz night:
Fri catch-up. Sam: venue quote still not in, will chase. Budget max 2k agreed. Leah to do flyer draft, maybe by Weds? Need someone to book the van. Raj said he’d ask his cousin about the PA system.
A good answer from the prompt above would look like this:
| Action | Owner | Due date | Source line |
|---|---|---|---|
| Chase the venue quote | Sam | Not agreed | “venue quote still not in, will chase” |
| Draft the flyer | Leah | Wednesday (unconfirmed) | “maybe by Weds?” |
| Book the van | Not agreed | Not agreed | “Need someone to book the van” |
| Ask about borrowing a PA system | Raj | Not agreed | “ask his cousin about the PA system” |
Decisions: budget capped at 2,000. Open questions: who books the van, and is Wednesday firm for the flyer?
Notice what the table does for you: three of the four actions have no agreed date, and one has no owner. Those gaps are exactly what to sort out before the next meeting. If the AI had quietly written “Raj, by Friday”, nobody would have noticed until the van was not booked.
Watch out for
- Invented details. Chatbots sometimes fill gaps with confident guesses, known as hallucination. The Source line column is your defence: no source, no action.
- Dates that drift. “Weds” or “next week” can turn into the wrong date. Check against a calendar.
- Bossy wording. AI can turn “Leah might draft the flyer” into “Leah must deliver the flyer”. Soften it before you send.
- Very long transcripts. A long recording may not fit in what the chatbot can read at once (its context window), and parts can be skipped. Split it in halves and check both.
- Privacy. Never paste passwords, bank details, health, disciplinary or HR matters, or anything a client has asked you to keep private. On free plans, assume what you paste may be stored.
Go deeper
- Explainer: Why AI models hallucinate
- Explainer: What is a large language model?
- Glossary: hallucination, context window, large language model
- Using AI: How to summarise a long PDF with an AI chatbot, without oversharing
- Tools directory: ChatGPT, Claude, Google Gemini, Microsoft Copilot
Sources
- Google Meet Help, “Take notes for me in Google Meet” (checked 19 Nov 2026)
- OpenAI Help, Data Controls FAQ
- Anthropic Privacy Center
- Google, Gemini Apps Privacy Hub



