You sit through 25 meetings a week, scribble notes in 5, send action items from 2, and forget the rest. This tutorial replaces that fog with a 5-minute post-meeting workflow that turns any Zoom, Meet, or Teams transcript into a clean Decisions / Action Items / Parking Lot doc. It is written for managers, PMs, consultants, and founders who run more meetings than they write notes for.
TL;DR
- The split that works: let the video platform (or a dedicated app) be the recorder; let ChatGPT or Claude be the analyst. Don’t ask one tool to do both well.
- Free path: Zoom AI Companion (free on any paid Zoom Workplace plan), Google Meet, or Teams export a transcript; paste it into ChatGPT or Claude with a strict extraction prompt. A 60-minute meeting is roughly 8,000–12,000 words, which fits in one prompt on every current flagship model.
- Paid path (better archive): Otter.ai Pro ($8.33/mo annual), Fireflies.ai Pro ($10/user/mo annual), or Granola ($14/user/mo) auto-capture and keep a searchable cross-meeting archive. Pricing as of June 2026.
- The one rule that prevents disasters: spot-check speaker attribution before you send anything. Misattributing an action item is the single failure that erodes trust fastest.
What this covers
A reusable workflow that takes a raw transcript to decisions and owned action items, plus a spot-check loop that catches the AI’s most common error: confusing who said what. We treat AI as the analyst, not the recorder. The recorder is your platform’s built-in transcription; the analyst is ChatGPT or Claude reading that transcript.
Who this is for
Anyone in 5 or more meetings a week — managers, PMs, consultants, founders, and engineering leads who run standups, planning, and design reviews. If you already take perfect notes by hand, this is overhead. If you regularly walk out of a meeting unsure what was decided, this changes your week.
When this is NOT the right tool
- Highly confidential meetings (compensation, M&A, personnel) where transcripts cannot leave the platform — handle by hand.
- One-on-one informal chats — the setup overhead exceeds the value.
- Meetings with no clear decisions or actions to extract — the AI will invent some, which is worse than nothing.
Pick your tools: recorder + analyst
You need two layers. The recorder produces a transcript; the analyst extracts structure from it. The recorder can be your meeting platform itself, or a dedicated note-taker that adds a searchable archive across every meeting.
Recorders (transcript source)
| Tool | Free tier (June 2026) | Paid entry | Captures by |
|---|---|---|---|
| Zoom AI Companion | Included on any paid Zoom Workplace plan; not on free Basic | Bundled, no extra fee | Native in-meeting (also joins Meet/Teams) |
| Google Meet / Teams | Transcript export on most paid Workspace/365 tiers | Bundled | Native in-meeting |
| Otter.ai | 300 min/mo, 30-min per conversation cap, 3 lifetime file imports | Pro $16.99/mo ($8.33/mo annual), 1,200 min/mo | Bot joins call |
| Fireflies.ai | 800 min storage, one-time AI credit pool | Pro $18/mo ($10/user/mo annual), 8,000 min storage | Bot joins call |
| Granola | Unlimited meetings, limited history | Business $14/user/mo | Captures device audio, no bot; audio deleted after transcription |
Granola’s no-bot model is worth calling out: it records your computer’s audio output directly, so there’s no “OtterPilot has joined” announcement in the room, and it deletes the audio after transcription rather than storing recordings. Zoom AI Companion is the cheapest serious option if your org already pays for Zoom — it’s bundled at no extra cost and already produces a summary and action items in the host’s post-meeting email.
Analysts (the extraction step)
For the paste-and-extract step, any current flagship handles a normal meeting:
| Model | In-app context for a pasted transcript | Notes |
|---|---|---|
| Claude Sonnet 4.6 / Opus 4.7 | 1M tokens at standard pricing | Strong at structured extraction; consumer plans now train on chats by default (see below) |
| Gemini 3.1 Pro | 1M tokens (Google AI Pro, $19.99/mo) | Good for very long, multi-session transcripts |
| ChatGPT (GPT-5.5) | Plus in-app ~320 pages; full 1M-token window only on the $200 Pro plan | Most convenient if you live in ChatGPT already |
A 60-minute meeting is roughly 8,000–12,000 words, which often lands above 20,000 tokens once speaker labels and timestamps are included. That still fits comfortably in a single prompt on any of the above. You only need to chunk when a transcript runs past about two hours.
Set a privacy policy first
This is the step most teams skip and later regret. As of June 2026, both ChatGPT and Claude consumer plans are eligible to train on your conversations by default — you must opt out.
- ChatGPT: turn off Settings → Data Controls → “Improve the model for everyone,” or use a Temporary Chat (never used for training, deleted after 30 days). ChatGPT Enterprise and Team are not trained on by default.
- Claude: turn off Settings → Privacy → “Improve Claude for everyone.” Leaving it on lets Anthropic retain conversations for up to 5 years; opting out returns you to the standard 30-day retention. Claude for Work (Team/Enterprise) and the API are not used for training by default.
If a transcript contains customer names, financials, or HR data, paste it only into a no-training environment: ChatGPT Enterprise/Team, Claude for Work, the API with retention disabled, or a self-hosted model. For one-off sensitive meetings, ChatGPT’s Temporary Chat is the fastest safe option.
Before you start
- Confirm your platform records and transcribes. Zoom, Meet, Teams, and Webex all do; some require the host to enable transcription explicitly. Test on one meeting first.
- Pick a single output schema for the team. “Decisions / Action Items (Owner, Due) / Parking Lot / Open Questions” covers 90% of meetings. Consistency makes the archive searchable.
- Pre-label speakers if the platform allows. Otter, Fireflies, and Read all support manual speaker labels; that one step removes the most common AI error.
Step by step
- Record the meeting on your platform (Zoom AI Companion, Meet, Teams) or via a dedicated app (Otter, Fireflies, Granola). On Zoom, the host’s post-meeting summary email already contains a first-draft summary and action items.
- Get the transcript. Most platforms export TXT or VTT; dedicated apps email it within a few minutes of the meeting ending.
- Open ChatGPT or Claude and paste the transcript with this strict extraction prompt:
Extract from this meeting transcript:
1. Decisions: only items explicitly agreed by 2+ participants.
2. Action items: bulleted as "owner — task — due date". If no due date
was stated, write "TBD".
3. Parking lot: items raised but explicitly deferred.
4. Open questions: items raised but not resolved.
Do not paraphrase or invent. Quote the exact line for each decision.
If a speaker attribution is unclear, mark it "unclear".- Review the output. AI occasionally confuses speakers and occasionally promotes a suggestion into a decision. Spot-check the 1–2 most important action items by searching the transcript for keywords.
- Send action items per owner. A Slack DM or email to each owner beats a group post every time. The owner who never sees their action does not do it.
- Archive the decisions doc somewhere persistent and searchable — Notion, a project channel, or a shared drive. Pick one and use it every time. Future-you will search this in six months.
First-run exercise
- Pick a low-stakes meeting first — a team standup or working session, not a board meeting. You’ll catch the AI’s mistakes faster on familiar material.
- Run the workflow once end-to-end without changing the prompt mid-flight. Save the output.
- Open the transcript and diff it against the AI’s output. Count: how many decisions are correct, how many action items have the right owner, how many were invented.
- On the second run, change only the prompt — add the rules you wished the first run had. After three iterations, the prompt is permanent.
Quality check before you send
- Every action item has an explicit owner. “The team will follow up” is not an action item; “Maria will email customer X by Friday” is.
- Every decision is one a participant could repeat aloud at the next meeting. If it’s softer than that, it’s a suggestion, not a decision.
- Speakers marked “unclear” are verified before sending. Misattributing an action item creates trust damage that takes weeks to repair.
- Parking-lot items have a next-step owner, even if the next step is “Pat brings this to next week’s planning.”
How to reuse this workflow
- Save the schema prompt as a saved instruction or a custom GPT. New meeting, same prompt, swap the transcript.
- For weekly recurring meetings, also feed in last week’s decisions doc as context. The model picks up continuity (“this resolves last week’s open question on pricing”).
- Quarterly, audit your decisions docs. Decisions that never got executed and action items stuck at “TBD” are the real meeting failures — and they’re obvious in the archive.
Adjacent workflows
- When the input is a long PDF (board deck, memo, regulatory filing) rather than a transcript, the Gemini PDF summarization workflow handles PDFs more reliably.
- When the meeting is a research-paper discussion you must walk into prepared, the 10-minute research-summary workflow gets you conversational on a paper without reading it cold.
- For on-the-go pre-meeting prep, the ChatGPT voice workflow covers rehearsing likely objections before you enter the room.
Common mistakes
- Trusting AI on speaker attribution without checking. Misattribution erodes trust faster than missing an item entirely.
- Asking for “meeting notes.” Too vague — ask for the four specific outputs above.
- Not following up on action items. The summary is the easy part; the chase is where projects move.
- Pasting confidential transcripts into a default consumer chat. Both ChatGPT and Claude train on consumer chats by default now — opt out or use a no-training tier.
- Letting AI infer decisions that were only suggestions. “Should we…” is not “We will…”.
- Not labeling speakers. Auto-labels are usually right for ~80% of utterances and badly wrong for the other 20%.
FAQ
- Which AI tool is best for this?: For convenience, ChatGPT or Claude with a pasted transcript. For privacy-sensitive content, use a no-training tier (ChatGPT Enterprise/Team, Claude for Work, or the API) — not a default consumer plan, since both now train on consumer chats unless you opt out.
- Do I need a dedicated meeting AI?: No, if you have Zoom AI Companion, Meet, or Teams transcripts plus ChatGPT or Claude. Dedicated tools (Otter, Fireflies, Granola, Read) save time on export and add a searchable cross-meeting archive. Cheapest serious option: Zoom AI Companion is free on any paid Zoom Workplace plan.
- How long a transcript fits in one prompt?: A 60-minute meeting is ~8,000–12,000 words. Claude Sonnet 4.6/Opus 4.7 and Gemini 3.1 Pro both take 1M tokens, so they swallow several hours in one go. ChatGPT Plus handles a single meeting fine in-app; only multi-hour transcripts need chunking. Past ~2 hours, summarize in 30-minute chunks then merge.
- What about multi-language meetings?: Current models handle mixed English/Chinese transcripts well, but state your output language explicitly in the prompt (e.g., “summarize in English”).
- Should I share the raw AI summary?: Always edit before sending. Raw AI output reads as careless even when it’s correct, and the edit pass is where you catch a misattributed owner.
- What if no one followed up on last meeting’s actions?: Open the next meeting with last week’s action list. Public accountability is the only reliable fix.