---
title: How to Use AI Meeting Notes and Pull Out Action Items
url: "https://goremy.ai/blog/articles/use-ai-to-take-meeting-notes-and-pull-out-action-items"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-11-12T22:00:18.878Z"
updated: "2026-11-12T22:00:18.878Z"
topic: Meeting Notes and Agendas
description: Learn how to use AI to record, transcribe and summarize meetings, extract action items with owners and dates, and write better notes for sales calls.
---

# How to Use AI Meeting Notes and Pull Out Action Items

> Learn how to use AI to record, transcribe and summarize meetings, extract action items with owners and dates, and write better notes for sales calls.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-12

## How do you use AI to take meeting notes?

You get a transcript of the meeting (from a recording, a built-in meeting tool, or a dedicated AI note taker), then give that transcript to an AI model with a specific prompt that asks for decisions, action items with owners and due dates, and open questions. Then you check the output against the transcript and send it out the same day. The AI does the typing and sorting. You still own accuracy, because it will sometimes get a name, a number or a commitment wrong.

The rest of this post gives you the steps, the exact prompts to paste (including one for sales calls), and a list of what to check before you hit send.

## TL;DR

- **Get a transcript first.** Good notes depend on a good transcript, so use a tool that records and transcribes the call (Zoom, Microsoft Teams and Google Meet all have AI features on some plans) or upload a recording to a separate note-taking tool.
- **Ask for a structure, not a "summary".** A prompt that demands decisions, action items with owner and date, and open questions produces something you can send. "Summarize this meeting" produces a paragraph nobody reads.
- **Check every owner, date and number against the transcript.** These are the details AI most often gets wrong, and they are the ones that matter.
- **For sales calls, extract buyer facts, not a recap.** Pain, budget, decision process, timeline, competitors, objections and agreed next steps are what your CRM and your manager need.
- **Get consent and follow company policy.** Tell people the meeting is being recorded, and do not paste confidential transcripts into an AI tool your company has not approved.
- **Move action items out of the notes.** A list inside a document gets forgotten. Put each item where you track work.

## What do you need before you start?

Three things, and the third trips people up.

1. **A way to capture the meeting.** Options are covered in the next section.
2. **An AI model you are allowed to use with meeting content.** Many companies approve a specific assistant (a Microsoft, Google or enterprise ChatGPT or Claude account, for example) and block personal accounts for work data. Ask your IT or security team if you are not sure. Meeting transcripts often contain customer names, pricing, hiring discussions and unannounced plans.
3. **Permission from the people in the room.** In many places, recording a call requires everyone's consent, and laws differ by state and country. The safe habit is to say it out loud at the start ("I'm recording this so I can write up notes, is that okay?") and to check your company's policy for customer calls.

## Which AI meeting notes tool should you use?

There are three broad categories. The right one depends mostly on what your company already allows.

| Option | Examples | Good for | Watch out for |
|---|---|---|---|
| Built-in meeting AI | Zoom AI Companion, Microsoft Teams with Copilot, Google Meet's note-taking feature | Internal meetings; already approved by IT in many companies | Often limited to certain paid plans; the summary format may not be what you want |
| Dedicated AI note takers | Otter, Fireflies, Fathom and similar tools | Joining many meetings across platforms; searchable archives | Bots that join the call are visible to guests; some customers object; check security review |
| Sales conversation tools | Gong, Chorus and similar tools | Sales and customer success teams who want call libraries, CRM sync and coaching | Usually bought by the team or company, not by an individual |

Features and plans change often, so check each vendor's current page rather than trusting a comparison written months ago (including this one).

If none of those are available to you, a plain recording plus a transcript works. Most meeting platforms can save a recording and a transcript file. You can paste the transcript text into an approved AI assistant.

## How do you get AI to write good meeting notes? (Step by step)

### Step 1: Prepare the meeting

Before the call, write two lines in your calendar invite or a notes document: the purpose of the meeting and the decision you need from it. Then paste these into your prompt later. AI works better with context. A transcript alone does not tell it that "the Q3 thing" means the budget reforecast.

Also write down the list of attendees with their roles. Transcripts often mislabel speakers, and a short attendee list lets the AI (and you) match names correctly.

### Step 2: Capture the transcript

Start the recording and transcription with the participants' knowledge. After the call, export the transcript as text. If your tool already produces a summary, still grab the full transcript. You will need it for checking, and you can run your own prompt on it to get a format that fits your team.

### Step 3: Run a structured prompt

Paste this into your approved AI assistant, then paste the transcript below it. Fill in the brackets.

```
You are helping me write notes for a work meeting. Use only what is in the transcript below. If something is not stated, write "Not stated" instead of guessing.

Meeting purpose: [one line]
Attendees and roles: [Name (role), Name (role)]
Date: [date]

Produce:
1. Summary: 3 to 5 sentences on what was discussed and concluded.
2. Decisions made: a bullet list. For each, note who made or approved it.
3. Action items: a table with columns Owner, Task, Due date, and Source (a short quote from the transcript that supports it). Only include tasks someone clearly agreed to do. If the owner or date was not stated, write "Not stated".
4. Open questions: things raised but not resolved, with who was asked to follow up if known.
5. Risks or concerns raised.

Keep the language plain. Do not add recommendations of your own.

Transcript:
[paste transcript]
```

Two parts of this prompt matter most. "Use only what is in the transcript" cuts down on invented details. The "Source" column forces the AI to point at the line it relied on, which makes checking fast.

### Step 4: Check the output against the transcript

This takes five minutes and is the step that protects your credibility. Go through this list:

- **Owners.** Did the person named actually agree to the task? AI often assigns a task to whoever spoke last about it.
- **Dates.** "By end of next week" is easy to misread. Convert relative dates to real dates, using the meeting date.
- **Numbers.** Prices, percentages, headcount, quantities. Compare each with the transcript.
- **Decisions versus ideas.** "We could try X" is not "We decided on X". AI tends to promote suggestions to decisions.
- **Missing items.** Search the transcript for phrases like "I'll", "can you", "let's", "next step" and "by Friday" and see whether each one appears in your action list.
- **Tone and sensitivity.** Remove anything that was said casually and should not be written down.

### Step 5: Send it quickly

Notes sent within an hour or two get read and corrected. Notes sent three days later do not. Use a second prompt to draft the email:

```
Turn the notes below into a short follow-up email to the meeting attendees. Start with a one-sentence summary, then list decisions, then action items grouped by owner with due dates. Ask people to reply by [date] if anything is wrong. Keep it under 200 words. Friendly and direct tone.

Notes:
[paste checked notes]
```

### Step 6: Move action items to where work lives

A list in an email is not a tracking system. Copy each action item into whatever your team actually uses: a task manager, a project board, your CRM, or a shared spreadsheet with columns for owner, task, due date and status. Ask the AI to format the list as CSV or a table you can paste straight in. Then review the open items at the start of the next meeting, which is when most follow-through gets decided.

## How do you use AI for sales call notes?

Sales calls need different notes than internal meetings. A recap of "what we talked about" is not useful. What helps is a record of what you learned about the buyer and what was agreed to happen next. This applies to account executives, customer success managers and anyone who talks to customers and has to update a CRM.

Use this prompt on a call transcript:

```
You are helping a salesperson write notes after a customer call. Use only what is in the transcript. If something is not stated, write "Not discussed". Do not infer budget, authority or timeline.

Account: [company]
Contacts on the call: [name, title]
Stage of the deal: [for example discovery, demo, negotiation, renewal]
Our attendees: [names]

Produce:
1. One-paragraph call summary.
2. Customer's problem or goal, in their own words where possible (use short quotes).
3. Current solution or process, and what they dislike about it.
4. Budget: anything said about money or approval limits.
5. Decision process: who decides, who else is involved, steps and timing.
6. Timeline: any dates or deadlines the customer mentioned.
7. Competitors or alternatives mentioned.
8. Objections or concerns, and how we responded.
9. Next steps: a table with Owner (customer or us), Task, Date.
10. Suggested CRM fields: next step, close date if stated, and any risks.
11. A draft follow-up email to the customer, under 150 words, that confirms next steps.

Transcript:
[paste transcript]
```

A few notes on using it well:

- **"Do not infer" matters more here.** An AI summary that guesses the customer has budget approved can end up in your forecast. If the transcript does not say it, the note should say "Not discussed", and that tells you what to ask on the next call.
- **Keep customer quotes.** Short exact quotes are the best material for proposals, internal handoffs and executive updates. Check them against the transcript, because AI sometimes tidies wording.
- **Customer success calls.** Swap items 4 to 7 for: health signals, feature requests, risks to renewal, who the champion is, and who is new. The structure stays the same.
- **Use the same prompt every time.** Consistent fields make notes comparable across deals, which helps managers and whoever picks up the account later.
- **Check your tool's CRM sync.** Dedicated sales tools can push notes into the CRM automatically. Review what they write, since a wrong field in the CRM is harder to spot later than a wrong line in an email.

## What about marketing, product and other meetings?

The same method works with a different set of fields. Change item 3 or the list of outputs to match the meeting:

- **Marketing:** campaign decisions, owners per deliverable, launch dates, approval status, budget changes.
- **Product:** decisions on scope, user feedback mentioned, bugs or requests raised, trade-offs discussed, who owns follow-up.
- **One-on-ones:** topics, commitments from each side, development goals. Keep these private and share only what both people agree to.
- **Interviews:** observations tied to specific answers. Be careful here. Do not let AI score candidates or draw conclusions, and follow your hiring process and legal guidance.
- **Customer research calls:** themes, quotes, pain points, and surprises.

## Where should you not trust AI meeting notes?

AI note takers are good at structure and bad at a few specific things. Know them.

- **Who said what.** Speaker labels get mixed up, especially with several people, accents or poor audio. Attribution matters for commitments.
- **Names, products and jargon.** Company names, acronyms and people's names are often transcribed wrong. Fix them before sending.
- **Numbers.** Amounts, dates and percentages are the highest-risk detail.
- **Implied agreements.** If someone said "sure, maybe" and the AI wrote "agreed", you have a problem. Look for hedging language.
- **Sarcasm, tension and silence.** A summary will flatten the tone. If the customer sounded unhappy, the notes will not say so.
- **Sensitive meetings.** Legal discussions, HR issues, performance reviews, and anything under confidentiality should not go through an AI tool unless your company has explicitly approved it for that use.
- **Completeness.** Long meetings produce summaries that skip things. The action items list is where missing something hurts most.

A good rule: AI drafts, you approve. Never forward AI notes you have not read.

## What are common mistakes with AI meeting notes?

- **Sending the raw AI summary.** Readers can tell, and errors reflect on you.
- **No agenda or context.** The AI cannot know what mattered most to you.
- **Skipping consent.** One surprised customer can cost more than the time saved.
- **Pasting confidential content into a personal AI account.** This is a policy violation in many companies, even when it feels harmless.
- **Letting action items stay in the document.** Without a place to track them, they disappear.
- **Treating the transcript as the notes.** A transcript is raw material. Nobody wants to read it.

## Can you build your own action item tracker?

Most meeting tools stop at the summary. The harder part is keeping action items visible after the meeting: who owes what, what is overdue, and what came out of which meeting. Many people end up with a spreadsheet that falls out of date. With [Remy](https://goremy.ai), you can describe a small internal tool in plain English, such as an action item tracker where you paste meeting notes, each item gets an owner, due date and status, and owners get an email reminder when something is overdue. Remy builds it as a full-stack app with sign-in and a database, and deploys it to a live URL. It costs a monthly subscription plus AI usage at cost, so it makes more sense for a team that runs many meetings than for occasional use.

## Frequently Asked Questions

### Is it legal to record a meeting and use AI to transcribe it?

It depends on where the participants are. Some places require only one party's consent, others require everyone's. Many companies also have their own rules for customer calls. The safe approach is to tell everyone at the start that you are recording and transcribing, and to check your company's policy or ask legal if you are unsure.

### Can I use ChatGPT or Claude for meeting notes?

Yes, if your company allows it for work content. Paste in a transcript with a structured prompt and you can get good results. Use the account your company approves, since consumer accounts may handle your data differently than enterprise ones. If you are not sure what is allowed, ask IT before pasting anything from a customer call.

### How accurate are AI meeting notes?

Usually accurate on the general flow of a conversation and less reliable on details. Names, numbers, dates, who agreed to what, and anything said with hedging are the usual problems. Audio quality, accents and overlapping speakers make it worse. Plan to check every action item against the transcript.

### What is the best way to get action items out of a meeting?

Ask for them in a table with owner, task, due date and a supporting quote from the transcript, and tell the AI to write "Not stated" when information is missing. Then confirm each item with the person named, send the list to attendees within a few hours, and move the items into the system where you track work.