How to Pull Action Items out of Any Meeting Recording with AI

Meeting action items are the concrete tasks a meeting produces, each with one owner, a due date and a clear definition of done. To pull them out of a recording with AI: transcribe the recording with speaker labels, ask a model to list every commitment as task, owner, due date and timestamp, then spend five minutes checking that list against the audio before it goes to the team. The steps, a tracker format, a prompt you can copy and a follow-up email template are below.

What is an action item in a meeting?

An action item is a commitment someone made out loud that has to turn into work after the call. It differs from the other two things meetings produce: a decision ("we launch in March") settles a question, a discussion point ("pricing feels high") records an opinion, and an action item ("Priya sends the revised pricing sheet by Thursday") assigns work. Meeting minutes usually contain all three; the action items list is the part people actually reopen.

A usable action item has four parts:

• A verb-first task that describes a finished result, not an activity ("Send signed contract to legal", not "Work on contract").

• One named owner. A team, "we" or "someone" is not an owner. Others can help; one person answers for it.

• A due date. A calendar date, not "ASAP" or "next sprint-ish".

• Context: the decision it comes from, or a link to the moment in the recording where it was agreed.

Action items examples: weak vs strong

Most bad action items fail in one of a few predictable ways. Here is how to rewrite them:

Action items template: a tracker you can copy

Keep action items in one flat list, one line per task. Five columns are enough; the fifth, the source timestamp, is what makes AI-extracted lists trustworthy, because anyone can jump to the exact moment the commitment was made.

ACTION ITEMS - [Meeting name], [date]
Recording / transcript: [link]

#  | Task (verb first, finished result)         | Owner  | Due date   | Status      | Source
---|--------------------------------------------|--------|------------|-------------|-------
1  | Send revised proposal with new pricing     | Maria  | 2026-11-19 | Open        | 12:04
2  | Share churn breakdown by plan in #growth   | Tom    | 2026-11-23 | Open        | 27:41
3  | Ship fix for empty-workspace error         | Lena   | 2026-11-18 | In progress | 33:10
4  | Ask legal if DPA covers subcontractors     | Priya  | 2026-11-24 | Blocked     | 41:55

Status values: Open / In progress / Blocked (say by what) / Done / Dropped (say why)

DECISIONS (context for the items above)
- Launch date stays 15 March (18:30)
- Annual plan discount drops from 25% to 20% (22:15)

OPEN QUESTIONS (not yet anyone's task)
- Do we need a separate EU onboarding flow? (38:02)

The same columns work in a spreadsheet, a Notion database or a Jira import. If you also keep formal minutes, this block slots straight into the "Action items" section of the AI meeting minutes templates we published separately.

How to get action items from a meeting recording with AI

The pipeline below works for any recording: a Zoom or Teams cloud recording, a voice memo from an in-person meeting, a phone call, or a webinar from last year.

1. Record and say so. Use the platform's native recording, or a phone in the middle of the table for in-person meetings. Announce it at the start; consent rules differ by country and US state, and our guide to recording consent in the US and EU covers them.

2. Transcribe with speaker labels. Upload the file to a transcription tool with speaker diarization, then rename Speaker 1, 2, 3 to real names. This is the step that makes owners extractable: "I'll send it Friday" is only an action item if the transcript knows who "I" is.

3. Keep timestamps in the text. Export or paste the transcript with timestamps so the model can cite them in the Source column.

4. Run the extraction prompt (below) on the full transcript. Ask for action items only, not a general summary; a focused request catches more of the small commitments made in passing.

5. Review against the audio using the checklist in the next section. Five minutes, not fifty.

6. Send the list and move it into your task system the same day.

A copyable prompt for AI meeting action items

This works in ChatGPT, Claude, Gemini or any assistant that accepts long text. Paste it, then paste the transcript underneath.

You are extracting action items from a meeting transcript.
The transcript has speaker names and timestamps.

Rules:
1. An action item is a commitment to do something after the meeting.
   Include requests that someone accepted ("Can you...?" "Sure").
   Exclude ideas, wishes and "we could someday" remarks.
2. For each item give: task (verb first, describes the finished result),
   owner (one person, exactly as named in the transcript),
   due date (as stated; if none was stated write "NOT SET"),
   source timestamp, and the exact quote it comes from.
3. If no single owner was agreed, write "NO OWNER" - do not guess.
4. Do not merge separate commitments into one item.
5. After the table, list separately:
   - Decisions made (with timestamps)
   - Open questions that nobody took ownership of

Output the action items as a table:
| # | Task | Owner | Due date | Timestamp | Quote |

Meeting date (to resolve "Friday", "next week"): [YYYY-MM-DD]

Transcript:

Two details in the prompt do most of the work. Asking for the exact quote lets you verify each item in seconds. Telling the model to write "NO OWNER" and "NOT SET" instead of guessing turns the gaps into a visible to-do for you rather than a confident invention. Fill in the meeting date before you run it, otherwise "Friday" has nothing to anchor to.

If you would rather not paste transcripts into a chat assistant, the transcript summarizer in BlaBlaScribe produces key points, decisions and action items from the same file you transcribed. AI summaries are on paid plans, and the free plan covers 30 minutes of transcription once; transcription runs on servers in Germany and files are never used to train AI. It works on uploaded recordings and links, so it does not join your calls as a bot.

How do Zoom, Teams, Google Meet, Otter and Fireflies show action items?

If your meetings already run on one of these platforms, you may get an action items draft without any extra tool. What each one offers, according to its own documentation:

All five need the meeting to happen on their platform or with their bot in the call. For in-person meetings, phone calls, external platforms you do not control, or old recordings, the record-transcribe-extract route above is the one that works everywhere. Our comparison of AI meeting note takers and the Otter vs Fireflies breakdown go deeper on the bots themselves.

How to review AI-extracted action items

Whatever produced the draft, a person who attended should review it before it goes out. AI is good at spotting commitment language ("I'll", "can you", "let's have X by") and weaker at judging whether a remark was a real commitment. Use this checklist:

1. Every item has one owner. Resolve each "NO OWNER" by naming someone or deleting the item.

2. Every item has a date. Replace "NOT SET" with a date the owner would agree to. If you are guessing, mark it as proposed in the email.

3. Delete the phantoms. Brainstorm remarks ("maybe we could redo the pricing page") often get extracted as tasks. If nobody said "I will", it is an idea, not an item.

4. Listen to anything surprising. Jump to the timestamp; checking a quote takes ten seconds with a timestamped transcript.

5. Look for what is missing. Quiet agreements ("yep, I've got it") and commitments made in the last two minutes, when people talk over each other, are the ones models miss most often.

6. Check names in mixed-language or noisy audio. If speakers were mislabeled, owners will be wrong; fix the labels first.

Action items follow-up email template

Send the reviewed list within a few hours, while everyone still remembers the conversation. Short beats complete: the full transcript is a link, not an attachment to read.

Subject: Action items - [Meeting name], [date]

Hi all,

Thanks for today. Here is what we agreed. Reply by [day] if anything
is wrong or a date doesn't work for you.

ACTION ITEMS
1. [Task] - [Owner] - due [date]
2. [Task] - [Owner] - due [date]
3. [Task] - [Owner] - due [date] (date proposed, please confirm)

DECISIONS
- [Decision]
- [Decision]

STILL OPEN (no owner yet)
- [Question] - we'll pick this up on [date]

Recording and transcript: [link]
Next check-in: [date and time]

[Your name]

Two habits make this email work. Name the owner in every line, so each person can find their own tasks at a glance. And ask for corrections by a set day: silence after that point counts as agreement, which settles "I never said I'd do that" before it comes up. For teams that work in several languages, translate the list rather than the whole transcript; it is short and the owners need to read it accurately.

Tracking action items until they are done

Action items lists fail after the email, not before it. A few practices keep them alive:

• Move items into the system where work happens on the same day: Jira, Asana, Linear, Trello, a shared sheet. Put the transcript timestamp link in the task description so the context travels with it.

• Review action items from the last meeting first. Open the next meeting with the previous list: two minutes, each owner says done, in progress or blocked. Teams that do this quickly learn that spoken commitments get checked.

• Use "Dropped" honestly. If an item no longer matters, mark it dropped with a reason. A list full of stale open items stops being read.

• Keep one running tracker per recurring meeting. For weekly meetings, append new items to the same list instead of starting a new document each week, so overdue items stay visible.

• Keep the transcript. The list drives the work; the transcript settles disputes about what was actually agreed weeks later.

The same pipeline applies to client calls, interviews, workshops and recorded phone calls; only the recording method changes. For platform-specific recording and export steps, see how to transcribe Zoom, Google Meet and Teams meetings.

Frequently asked questions

What are action items in a meeting?

Action items are the tasks a meeting assigns: specific pieces of work with one owner and a due date that must be done after the meeting. They differ from decisions, which settle a question, and from discussion notes, which record opinions.

How do you write action items from meeting notes?

Start each item with a verb that describes the finished result, name one owner, add a calendar due date and link to the decision or transcript moment it came from. "Maria sends the revised proposal to Acme by 19 November" is a complete action item; "follow up on proposal" is not.

Can AI extract action items from a meeting recording?

Yes. Transcribe the recording with speaker labels, then ask an AI model to list every commitment with owner, due date and timestamp. Expect to correct a few items: models catch commitment phrases well but sometimes list ideas as tasks or miss quiet agreements, so a five-minute review by someone who attended is part of the process.

What should an action items template include?

At minimum: task, owner, due date and status. Adding a source column with the transcript timestamp makes each item verifiable, and separate sections for decisions and open questions keep the task list clean.

Where do Zoom and Teams put meeting action items?

Zoom AI Companion includes next steps and action items in its meeting summary. Microsoft Teams shows AI-generated notes and follow-up tasks in the AI summary of the meeting recap, which needs Teams Premium or a Copilot license and a transcribed meeting.

How do you track action items from previous meetings?

Keep one running tracker per recurring meeting, move each item into your task tool the same day, and open the next meeting by reviewing last meeting's list item by item. Mark items done, blocked or dropped with a reason so the list stays accurate.