AAI Growth Lab
AI Workflows Beginner 15 minutes

Meeting → Notes → Action Items

Record any meeting or lecture, turn the transcript into clean structured notes, extract owned action items, and send the follow-up email — in under 15 minutes with Otter.ai, ChatGPT and Notion AI.

Difficulty
Beginner
Completion time
15 minutes
Time saved
45–60 minutes per meeting
Start Workflow Jump to prompt library

Workflow overview

The cost of a meeting is not the hour it takes. It is the week afterwards when nobody remembers who agreed to do what, and the same discussion happens again.

This workflow closes that gap. Otter.ai captures the conversation with speaker labels, ChatGPT converts the messy transcript into a structured record and an owned task list, and Notion AI files it where the team already works.

The whole loop runs while people are still walking back to their desks — which is the point. A follow-up email sent within an hour changes behaviour; one sent two days later gets archived unread.

Goal

Turn every meeting or lecture into a written record, a task list with owners and dates, and a follow-up email that goes out the same day.

Best for: Team leads, freelancers running client calls, and students who record lectures with permission.

Required AI tools

Otter.ai

Live transcription of meetings and lectures with speaker labels.

ChatGPT

The reasoning and writing engine at the centre of every workflow.

Notion AI

Turns raw notes into structured pages, tasks and databases.

The workflow at a glance

1
Record & transcribe
2
Structured summary
3
Extract action items
4
Log in Notion
5
Send follow-up

Step-by-step guide

Work through the steps in order. Every prompt is copy-and-paste ready — swap the bracketed parts for your own details.

1Record and transcribe with Otter.ai2Turn the transcript into structured notes3Extract action items with owners and dates4File everything in Notion5Generate and send the follow-up email
1

Record and transcribe with Otter.ai

Otter.ai During the meeting

Connect Otter to Google Calendar so it joins Zoom, Meet or Teams calls automatically, or use the mobile app for in-person meetings and lectures. Always announce that you are recording — in many places it is a legal requirement and it is good practice everywhere. After the call, open the transcript, fix any mislabelled speakers, and copy the full text.

Prompt

Setup checklist: 1. Otter > Settings > Calendar — connect your work calendar. 2. Enable auto-join for events with a video link. 3. After the call, open the transcript and rename speakers to real names. 4. Select all and copy the transcript, including speaker labels.

What you should get back

A timestamped transcript with named speakers, ready to paste into ChatGPT. A 45-minute meeting typically yields 6,000–9,000 words.

2

Turn the transcript into structured notes

ChatGPT 4 min

Paste the transcript and ask for a specific structure rather than a generic summary. Naming the sections you want is the difference between a paragraph nobody reads and a document the team refers back to. Ask it to preserve any numbers, dates and named commitments verbatim.

Prompt

You are a meticulous meeting scribe. Below is a transcript of [MEETING NAME] on [DATE] with [ATTENDEES]. Produce notes in exactly this structure: 1. One-paragraph summary (max 60 words) 2. Decisions made — bullet list, each with the reasoning given 3. Discussion points — grouped by topic, with any disagreement noted 4. Numbers, dates and commitments mentioned — quote them exactly 5. Open questions with no answer yet Rules: never invent detail; if something was ambiguous, write "unclear from transcript". Use plain language. Transcript: [PASTE TRANSCRIPT]

What you should get back

A one-page structured record: a short summary, 3–6 decisions with rationale, topic-grouped discussion notes, exact figures, and a list of unresolved questions.

3

Extract action items with owners and dates

ChatGPT 3 min

Run a second, separate pass purely for tasks. Splitting summarisation and extraction produces noticeably better results than asking for both at once, because the model is not trading detail against brevity. Insist on a table so it drops straight into a project tool.

Prompt

From the same transcript, extract every action item. Return a markdown table with columns: Task | Owner | Due date | Priority | Source quote Rules: - Owner must be a named attendee. If nobody was named, write UNASSIGNED. - If no date was given, infer a sensible one from the discussion and mark it (inferred). - Priority: High / Medium / Low, based on stated urgency and dependencies. - Source quote: the exact line where the commitment was made. - List UNASSIGNED rows at the top. Transcript: [PASTE TRANSCRIPT]

What you should get back

A table of 5–15 tasks. Unassigned items sit at the top so the organiser can chase an owner before publishing the notes.

4

File everything in Notion

Notion AI 4 min

Paste the notes into a Meetings database page and the action table into a linked Tasks database. Markdown tables paste natively in Notion and convert cleanly. Then use Notion AI on the page to link related past meetings and generate a short 'context for newcomers' block at the top.

Prompt

Notion AI prompt (run on the meeting page): Summarise this meeting in three bullets for someone who was not present, then list any topic here that also appeared in our previous meeting notes and note what changed since then.

What you should get back

A Meetings page with the full record, a linked task database populated with owners and dates, and a three-bullet context block at the top.

5

Generate and send the follow-up email

ChatGPT 4 min

The follow-up email is what makes the work visible. Keep it short: what was decided, who owes what by when, and one line on the next checkpoint. Read it once before sending — you are the one accountable for the contents, not the model.

Prompt

Write a follow-up email for [MEETING NAME] held on [DATE]. Audience: [ATTENDEES + anyone who missed it] Tone: warm, direct, professional. No corporate filler. Length: under 200 words. Structure: - One-line purpose reminder - "What we decided" — max 4 bullets - "Who is doing what" — name, task, date, one line each - "Next checkpoint" — date and format - Sign off inviting corrections by [DEADLINE] Use these notes and this action table: [PASTE NOTES + TABLE]

What you should get back

A ready-to-send email under 200 words, with decisions, owned tasks, a next checkpoint, and an explicit invitation to correct the record.

Copy-and-paste prompt library

Every prompt used in this workflow, in one place. Tap copy and paste straight into your tool of choice.

Prompt 1 — Structured meeting notes

You are a meticulous meeting scribe. Produce notes with: 60-word summary; decisions with reasoning; discussion points grouped by topic; exact numbers, dates and commitments; open questions. Never invent detail — write "unclear from transcript" where ambiguous.

Prompt 2 — Action item extraction

Extract every action item as a markdown table: Task | Owner | Due date | Priority | Source quote. Owner must be a named attendee or UNASSIGNED. Mark inferred dates. Put UNASSIGNED rows first.

Prompt 3 — Follow-up email

Write a follow-up email under 200 words: one-line purpose, what we decided (max 4 bullets), who is doing what with dates, next checkpoint, and an invitation to correct the record by [DEADLINE]. Warm, direct, no filler.

Prompt 4 — Lecture note variant

This is a lecture transcript for [COURSE]. Produce: key concepts with definitions, worked examples step by step, anything the lecturer flagged as examinable, and a list of terms I should look up. Use one short line per fact.

Prompt 5 — Decision log entry

From this transcript, write one decision log entry per decision: Decision, Date, Owner, Alternatives considered, Reason chosen, Reversal cost. Keep each field to one sentence.

Prompt 6 — Client recap

Rewrite these internal notes as a client-facing recap. Remove internal debate, keep commitments and dates, add a short paragraph confirming scope, and end with what we need from the client and by when.

Expected output examples

Decisions section

• Ship the pricing page redesign on 14 Aug — current page converts at 1.2% vs 3.1% on the old layout. • Hold the enterprise tier until Q4 — sales wants three more discovery calls first. • Keep annual billing discount at 20% — unclear from transcript whether finance signed off.

Action item table

| Task | Owner | Due | Priority | | Rewrite pricing copy | Maya | 4 Aug | High | | Pull conversion data | Tom | 1 Aug (inferred) | High | | Book 3 discovery calls | UNASSIGNED | 8 Aug | Medium |

Follow-up email

Subject: Pricing page — decisions and next steps Quick recap of today's call. What we decided: ship the redesign 14 Aug; hold the enterprise tier to Q4; keep the 20% annual discount pending finance. Who is doing what: Maya — new pricing copy by 4 Aug. Tom — conversion data by 1 Aug...

Notion context block

For anyone catching up: the pricing page has underperformed since the March redesign, this is the third session on it, and the difference from last time is that we now have segment-level conversion data.

Real-world example

Six-person agency, four client calls a week

A small design agency was losing roughly an hour per client call to note-writing, and about one commitment a month was slipping because it lived only in someone's memory.

They connected Otter to the shared calendar so every client call was transcribed automatically, and made the two-prompt pass a standard closing task for whoever ran the call.

Within a month, follow-up emails went out the same day instead of two days later, unassigned tasks were caught in the meeting itself because the table made them obvious, and the notes archive became the reference point for scope disputes. Time per meeting dropped from around 55 minutes of admin to 12.

If you're a student

With your lecturer's permission, record lectures on the Otter mobile app and run the lecture-note variant prompt. You end up with concept definitions, worked examples and a list of examinable points — which feeds straight into the flashcard workflow.

If you run a business

Make the follow-up email non-negotiable after every client call. A same-day written recap with owners and dates is the cheapest scope protection a small business can buy, and it makes you look considerably more organised than competitors.

Best practices

  • Always announce recording at the start of the call and get an audible yes.
  • Fix speaker labels before summarising — mislabelled speakers create wrongly assigned tasks.
  • Run summarisation and extraction as two separate prompts; combined prompts lose detail.
  • Require a source quote for every action item so anyone can verify it against the transcript.
  • Send the follow-up within an hour while the meeting is still fresh.
  • Never send AI-written notes unread — you are accountable for what they say.

Common mistakes

  • Pasting a raw transcript with no structure request

    "Summarise this meeting" returns a bland paragraph. Naming the exact sections you want is what makes the output usable.

  • Letting tasks stay unassigned

    An action item without a named owner will not happen. Chase the owner before you publish the notes.

  • Recording without consent

    Beyond the legal exposure in two-party-consent regions, it destroys trust with clients if discovered later.

  • Pasting confidential transcripts into free consumer tiers

    Check your data settings, or use a business plan with training disabled, before processing client or HR conversations.

Variations

Lecture mode

Swap the notes prompt for the lecture variant: concepts, worked examples, examinable points and unknown terms. Feed the result into the flashcard workflow the same evening.

Client-facing recap

Generate two versions from one transcript — a candid internal record and a polished client recap with internal debate removed.

Fully automated with Zapier

Trigger on a new Otter transcript, send it to ChatGPT via Zapier, create Notion rows for each action item, and draft the follow-up email in Gmail automatically.

Frequently asked questions

How accurate is Otter.ai on accented or technical speech?

Generally 85–95% on clear audio. Technical vocabulary and heavy background noise hurt it most. Fix key terms in the transcript before summarising — a wrong product name propagates through every downstream output.

Can I do this without paying for anything?

Yes. Otter's free tier covers a limited number of transcription minutes per month, ChatGPT's free tier handles the prompts, and Notion's personal plan is free. Heavy meeting loads eventually justify a paid transcription plan.

What if the transcript is too long to paste?

Split it into 20-minute chunks, summarise each, then run a final prompt over the combined summaries. Extract action items from the chunks, not the summaries, or you will lose commitments.

Is it legal to record a meeting?

It depends on your jurisdiction — some regions require consent from all parties. Announce recording at the start every time and stop if anyone objects; that satisfies both the rules and the room.

Can this replace a human notetaker?

For routine internal meetings, largely yes. For negotiations, performance reviews and legal discussions, keep a human in the loop — nuance and tone matter more than coverage.

Get the next workflow in your inbox

One practical AI workflow every week — no fluff, no hype, just the steps and prompts.

Previous workflowResearch → Summary → FlashcardsNext workflow Product → Landing Page → Ad Copy

Related workflows

Workflow

Research → Summary → Flashcards

Turn lecture material into a study-ready flashcard deck.

Open workflow Workflow

Customer Email → AI Reply → CRM

Answer customer email faster and log it automatically.

Open workflow Workflow

PDF → Summary → Presentation

Turn a long report into a polished deck with speaker notes.

Open workflow

Put this workflow to work today

Compare the tools used here, then go deeper with our written guides.

Browse AI Tools Read: Automating Customer Emails