Project: Meeting Note Summarizer
The average professional attends over 20 hours of meetings per month — and most of that content evaporates within 48 hours. AI changes this: a meeting can now be automatically transcribed, summarised, and turned into a structured action list in under two minutes. This project walks you through three approaches, from zero-setup built-in tools to a flexible manual pipeline that works for any meeting on any platform.
Three Approaches — Pick What Fits
| Approach | Best For | Setup | Cost |
|---|---|---|---|
| Built-in platform AI | Teams or Google Meet users — zero extra tools | Click a button during the meeting | Included in M365 Copilot / Google Workspace |
| Dedicated AI note-taker | Works across Zoom, Meet, Teams — fully automatic | Install app, connect calendar — 10 minutes | Free tiers available (Fathom, Otter.ai) |
| Manual transcript pipeline | Any meeting, any platform, full control over output format | Record → transcribe → paste prompt | Free (uses existing AI subscription) |
Approach 1 — Built-In Platform AI
If your organisation already uses Microsoft Teams or Google Meet, AI meeting notes may already be available to you — no extra tool, no extra cost.
Microsoft Teams Copilot
Teams Copilot listens during the meeting and can answer questions in real time ("What did we decide about the budget?") as well as generate a full structured recap after the call. The Recap tab automatically appears after a recorded meeting, grouping key discussion points, decisions, and action items — each with the owner's name and a suggested deadline where spoken.
What it does well
- Real-time Q&A during the call — "Catch me up on what I missed" if you join late
- Automatic Recap tab after the meeting with structured summary
- Action items with attributed owners and suggested deadlines
- Summarises up to 30 days of Teams chat content
- Integrates directly with Teams channels, SharePoint, and Outlook
Limitations
- Only works in Microsoft Teams — no Zoom, Google Meet, or other platforms
- Requires Microsoft 365 Copilot licence ($30/user/month on top of M365)
- Recording must be enabled for the meeting
Google Meet — "Take Notes for Me"
Google Meet's Gemini integration adds a "Take notes for me" button to any call. When enabled, Gemini listens throughout, then automatically generates a recap document — shared with all participants — containing discussion highlights, key decisions, and suggested next steps. As of March 2025, this works in English, French, German, Italian, Japanese, Korean, Portuguese, and Spanish.
How to use it:
- Start or join a Google Meet call
- Click the pencil icon (Take notes for me) in the meeting controls
- Gemini begins capturing — a small indicator shows participants that notes are being taken
- After the meeting, a Google Doc is automatically created and shared with all participants
Requires: Google Workspace Business Standard or higher with Gemini AI features (included in all Workspace Business plans at no extra charge since January 2025).
Approach 2 — Dedicated AI Note-Taker Tools
Dedicated AI note-takers work across any video platform — Zoom, Teams, Meet, Webex — and provide more control over output format, integrations, and searchability than built-in tools. They join your call as a participant (or, in the case of newer bot-free tools, capture system audio invisibly) and deliver structured notes after the call.
Fathom — Best Free Option
Fathom offers unlimited recordings on its free tier — making it the standout choice for anyone who wants a fully automatic note-taker without paying. It joins calls as a participant, transcribes in real time, and delivers an AI summary within 30 seconds of the call ending. It supports 28 languages and has recently added a botless recording option (beta) for when you do not want a visible bot participant.
Granola — Bot-Free, Device Audio
Granola takes a different technical approach: instead of joining your call as a visible bot participant, it captures audio directly from your device — invisibly, without appearing in the participant list. This makes it useful for sensitive client meetings, interviews, or situations where a visible bot would be awkward. It works with any video call platform that plays audio through your computer.
Granola emerged as the most-discussed AI note-taker in late 2025 communities specifically because of this bot-free approach. You take rough notes during the call as normal, and Granola enhances and structures them using the audio transcript after the meeting.
Otter.ai — Long-Standing, Feature-Rich
Otter.ai is one of the oldest AI transcription services and offers a broad feature set: real-time transcription, speaker identification, AI-generated summary, action item extraction, and integration with Zoom, Teams, and Google Meet. Its OtterPilot feature automatically joins scheduled calls from your calendar. Free tier is limited to 600 minutes per month of transcription.
Fireflies.ai — Enterprise-Grade, Multi-Language
Fireflies.ai positions itself as the most comprehensive meeting intelligence platform, with strong multi-language support (useful for global teams), CRM integrations (Salesforce, HubSpot), and a searchable library of all your meeting transcripts. It can identify action items, questions, and decisions across large transcript libraries and has conversation intelligence features for sales teams.
Approach 3 — Manual Transcript Pipeline
If you cannot use a bot tool (privacy policy, platform restrictions, no subscription) or want full control over your note format, the manual pipeline is the most flexible approach. It works for any meeting on any platform, costs nothing beyond your existing AI subscription, and produces exactly the output format you specify.
Step 1 — Get the transcript
If the platform supports it (recommended):
- Zoom: enable transcription in settings → after the meeting, download the .vtt transcript file from the Zoom web portal
- Teams: record the meeting → Teams generates a transcript automatically accessible in the meeting chat
- Google Meet: use "Take notes for me" to get a Google Doc, or enable transcription in Meet settings
If no built-in transcription (fallback):
- Record the audio (with all participants' consent) using your platform's record feature or a screen recorder
- Upload the audio file to ChatGPT (Advanced Voice/file upload), Claude, or a dedicated STT tool like Whisper or ElevenLabs Scribe
- Get the text transcript back
Step 2 — Clean the transcript (optional but helpful)
Auto-generated transcripts often contain filler words ("um", "uh"), repeated phrases, and speaker attribution errors. Before summarising, ask AI to clean it:
Step 3 — Summarise with a prompt
Paste the transcript and one of the prompt templates below. The output is immediately shareable with participants.
Ready-to-Use Summary Prompts
Copy the prompt that matches your meeting type. Replace the bracketed sections, paste the transcript at the end, and send. Output is ready to copy into an email or doc.
General meeting (works for 80% of meetings)
Summarise this meeting transcript into a structured note. Use exactly these sections:**Meeting:** [title] | [date] | [attendees]
**Summary:** 3–4 sentence overview of what was discussed and decided.
**Key Decisions:** Bulleted list — only firm decisions, not discussions.
**Action Items:** Table with columns: Action | Owner | Due Date. Only include items explicitly assigned or agreed.
**Open Questions:** Items raised but not resolved.
**Next Steps:** What happens next and when.
Keep it factual. Do not infer decisions that were not clearly made. Transcript: [paste here]
Executive / leadership meeting
You are a Chief of Staff summarising a leadership meeting for busy executives. Produce:**TL;DR:** Two sentences — what was decided and what happens next.
**Strategic Decisions:** What was agreed (not discussed — agreed).
**Risks & Issues Raised:** Problems surfaced that require attention.
**Commitments Made:** Who committed to what and by when.
**Items Deferred:** Topics raised but deliberately parked for later.
Tone: direct, no filler. Transcript: [paste here]
Project / team standup
Summarise this standup/team meeting. Produce:**Progress:** What was completed since last time (by person if attributable).
**Blockers:** Issues preventing progress — who has them, what they need.
**Today's Focus:** What each person or team is working on next.
**Action Items:** Owner | Task | Due.
**Decisions:** Any process or priority decisions made.
Keep each bullet to one sentence. Transcript: [paste here]
Client / sales meeting
Summarise this client meeting from our perspective. Produce:**Client Context:** Who we met with and what they care about.
**Their Key Needs / Pain Points:** What they said they need (direct quotes where possible).
**What We Agreed:** Commitments we made to the client.
**What They Agreed:** Commitments the client made.
**Our Next Actions:** What our team must do and by when.
**Relationship Notes:** Anything useful for future conversations (preferences, concerns, tone).
Transcript: [paste here]
1:1 / performance conversation
Summarise this 1:1 meeting as private notes (not for distribution). Produce:**Topics Discussed:** Brief bullet per topic.
**Feedback Given:** What feedback was shared and how it was received.
**Goals / Commitments:** What was agreed to work on or achieve.
**Support Needed:** What the person needs from their manager or team.
**Follow-up:** When we meet next and what to check on.
Transcript: [paste here]
What Works Well
Action item extraction
AI is very good at identifying explicit action items from meeting transcripts — things that were clearly assigned ("can you send that by Friday?", "I'll follow up with legal"). The output is often more complete than notes taken manually during the call, since the human note-taker is also trying to participate.
Instant shareable output
A structured meeting summary that would previously take 20–30 minutes to write up manually is ready to copy-paste and send within two minutes. For recurring meetings with consistent formats, this saves hours per month.
Catching things you missed
In any meeting where you were actively participating, you were not fully listening to every word. The transcript + AI summary catches commitments, questions, and context you were too busy contributing to register consciously. This is especially useful for long, complex meetings with many participants.
Searchable meeting archive
If you store your meeting summaries in a consistent location (a Notion database, a Google Drive folder, NotebookLM), you build a searchable archive of decisions made, commitments given, and issues raised. Tools like Fireflies.ai and Otter.ai build this automatically. "What did we decide about the pricing model in Q3?" becomes answerable.
What to Watch Out For
Always get consent before recording
Recording a meeting without the knowledge of all participants is illegal in many jurisdictions and a breach of trust in all of them. Before recording any call — with a bot, a screen recorder, or any other tool — inform all participants that the meeting is being recorded and for what purpose. Most platforms display a notification automatically when recording starts; never disable or obscure this. For external meetings with clients or partners, get explicit verbal or written consent.
AI invents action items that were not agreed
AI can extract items that were discussed as possibilities but not actually agreed — and present them as firm action items. Always review the action item list against your recollection before sending it to participants. The prompt template above specifies "only items explicitly assigned or agreed" — but still check. A fabricated action item attributed to someone who never agreed to it creates confusion and erodes trust.
Poor audio = poor transcript = poor summary
The quality of the AI summary is bounded by the quality of the transcript, which is bounded by the audio quality. Background noise, overlapping speakers, heavy accents, and poor microphones all degrade transcription accuracy significantly. For important meetings, use headsets, mute when not speaking, and ensure everyone is in a quiet environment. Review the raw transcript before summarising if the meeting was noisy or had difficult audio.
Nuance and tone are lost
A meeting transcript captures words but not tone, body language, or subtext. An AI summary of a tense negotiation where both sides were being polite but actually disagreeing strongly will look like a productive discussion. For meetings where the interpersonal dynamics matter as much as the content — performance conversations, difficult negotiations, sensitive personnel discussions — add your own tone notes that the AI cannot capture.
Confidential meetings and data privacy
Meeting transcripts often contain sensitive business information — financial figures, personnel matters, unreleased product plans, legal strategy. Check your organisation's policy on which AI tools are approved for use with confidential information before uploading a transcript anywhere. Consumer-tier AI tools (free ChatGPT, free Otter.ai) may train on uploaded content by default. Use enterprise tiers or dedicated on-premise solutions for genuinely sensitive meetings.
Tool Comparison
| Tool | Bot? | Platforms | Free Tier | Best For |
|---|---|---|---|---|
| Teams Copilot | Built-in | Teams only | No ($30/user/month add-on) | Microsoft ecosystem orgs |
| Google Meet Gemini | Built-in | Meet only | Included in Workspace Business | Google Workspace orgs |
| Fathom | Yes (bot-free beta) | Zoom, Teams, Meet | Yes — unlimited recordings | Free automatic note-taking across platforms |
| Granola | No — device audio | Any (system audio) | 25 lifetime meetings | Client meetings where bot is awkward |
| Otter.ai | Yes | Zoom, Teams, Meet | 600 min/month | Feature-rich platform, integrations |
| Fireflies.ai | Yes | Zoom, Teams, Meet, more | Limited storage | Global teams, sales, meeting archive |
Ways to Extend This Project
Auto-send notes to your team
After generating your summary, paste it into a standard email template and send it within one hour of the meeting. Consistent, fast note distribution builds a reputation for reliability and ensures commitments are visible before they are forgotten.
Build a searchable meeting archive
Store every meeting summary in a consistent location — a Notion database with date, meeting title, attendees, and summary as fields. After a few months you have a searchable record of every decision made. Ask AI to search it: upload several summaries to NotebookLM and query across them.
Generate follow-up emails automatically
After generating the summary, add one more prompt: "Based on these meeting notes, write a brief follow-up email to all participants. Subject line: 'Notes: [meeting name]'. Include a thank-you, the summary, action items table, and a warm close." Draft is ready in seconds — just review and send.
Create a weekly digest
At the end of the week, paste all five meeting summaries into one AI prompt: "I had these meetings this week. Produce a weekly digest: top 5 decisions made, open action items I own, open questions still outstanding, and anything that needs my attention next week." Turns fragmented weekly activity into a coherent management view.
What Is New in 2025–2026
Bot-free capture goes mainstream
The biggest trend in AI meeting notes in 2025 is the shift away from visible bot participants. Granola's device-audio approach and Fathom's bot-free beta reflect growing demand for invisible capture — particularly for client-facing and sensitive meetings where a named bot participant changes the conversation dynamic. Expect bot-free to become the default approach for most tools by 2026–2027.
Google Meet "Take Notes for Me" in 8 languages (March 2025)
Google's expansion of its built-in meeting notes to eight languages in early 2025 brought AI meeting notes to non-English-speaking Workspace users at scale — with no extra tool or cost. For organisations using Google Workspace, this is now effectively a free, built-in capability that most users have not discovered yet.
Real-time meeting intelligence
Teams Copilot's real-time Q&A ("catch me up on what I missed") is the leading edge of a new category: AI that participates in the meeting rather than just summarising it afterwards. Expect tools to surface relevant context, suggest talking points, and flag commitments in real time during the call — not just post-call.
Checklist: Do You Understand This?
- Can you describe the three approaches to AI meeting notes, and which you would choose for a confidential client call?
- Can you explain the legal and ethical requirement before recording any meeting, and what happens if you skip it?
- Can you describe what makes Fathom the best free option, and what makes Granola different from other tools?
- Can you walk through the three steps of the manual transcript pipeline for a meeting on a platform with no built-in AI?
- Can you explain the "AI invents action items" failure mode and what you should always do before sending the summary to participants?
- Can you describe the data privacy risk of uploading a meeting transcript to a consumer AI tool, and how to mitigate it?
- Can you name two ways to extend the basic meeting note project into something more valuable over time?
- Can you explain what "bot-free capture" means and why it is becoming the preferred approach for client meetings?