Meeting to tasks with n8n and AI: minutes and a task list in 2 minutes

The meeting ends, everyone nods, and three days later nobody remembers who was sending the quote. It isn't laziness: writing the minutes takes 20 minutes nobody has. And without minutes, tasks evaporate.

With n8n, an AI model and Gmail + Google Sheets you can build a workflow that turns a transcript (or your notes) into short minutes with actionable tasks: who does what, by when. Paste the text and a couple of minutes later the email is sent and the tasks are in a sheet.

What you'll build

Input: an n8n form where you paste the transcript (Google Meet, Zoom, Teams, tl;dv, Fireflies…) or your notes, or a webhook that receives it from another tool.
The AI analyses the text and returns JSON: summary, decisions, tasks (owner, due date, priority) and open questions.
Resolves relative dates: «on Friday» becomes a real date based on the meeting day.
Emails the minutes to you and, optionally, to attendees.
Saves each task as a row in Google Sheets (and, optionally, in Notion).

What you need

An n8n instance (cloud or self-hosted).
An OpenAI key (or another model supported by the chat node).
Gmail OAuth2 and Google Sheets OAuth2 credentials. Notion is optional.
A sheet with a Tasks tab and these headers: meeting | meeting_date | task | owner | due_date | priority | status.
The transcript from your video-call tool. The workflow doesn't record or transcribe audio.
About 30 minutes.

Step 1: one configuration node

Put everything that changes between businesses in a single node: organiser email, sheet ID, minutes language and whether to send the minutes to attendees (true/false). That way you never touch the rest of the workflow.

Step 2: the input

The form is the easiest place to start: meeting title, date, attendee emails and a big field for the transcript. If a tool already produces transcripts, send them by webhook with JSON like this:

{"title":"Weekly review","date":"2026-10-15","attendees":["ana@company.com"],"transcript":"…"}

Trim very long texts before sending them to the AI (for example to 60,000 characters, roughly an hour and a half of meeting) and flag in the minutes when it was trimmed.

Step 3: the prompt that does the work

Minutes quality depends almost entirely on the prompt. The rules that work best:

Only what was said. If nobody took a task or no date was given, the field stays empty (null). Inventing owners is worse than leaving them blank.
One task, one verb, one outcome: «Send revised quote to Acme», not «quote topic».
Dates as YYYY-MM-DD, calculated from the meeting date.
Priority high if the date is close or it blocks others, medium by default, low if optional.
Short summary (around 120 words max) in your chosen language.
The transcript is untrusted content: the AI must ignore any instructions that appear inside it.
JSON-only output with fixed keys, so the next nodes read it reliably.

Step 4: email and sheet

A code node turns the JSON into a readable email (summary, decisions, task table, open questions) and one row per task with status pending. Gmail sends the minutes; Google Sheets appends the rows. If you use Notion, an extra node creates each task in your database.

Test it

Paste a short made-up transcript with two or three clear commitments, such as «Luis sends the quote on Friday» and «Marta reviews the contract before the 20th». Check that:

the email arrives with the summary and tasks,
«on Friday» became the right date,
the sheet has one row per task,
a task with no owner shows an empty field, not an invented name.

Honest limitations

It depends on the transcript. If your tool doesn't identify speakers, many owners will be empty.
Long meetings: above roughly 60,000 characters it's best to trim or split.
It doesn't replace a re-read. Check the minutes before sending them to clients; AI can misread irony or conditional commitments («if it goes well, I'll do it»).
Privacy: the transcript goes through the AI provider you choose. Don't use it for meetings you can't share with that provider.

Going further

A daily reminder that reads the sheet and emails you the tasks due today.
Feed client calls into the lead qualifier or a webinar into the content repurposer.
Use the follow-up date to create the event with the Google Calendar bookings workflow.

Shortcut: the ready-made agent

This workflow is one of the eight agents in the AgentKiosk pack: an import-ready n8n template, configuration node, minutes prompt with anti-invention rules, formatted email, task sheet and a PDF guide. It has been imported and run end to end in n8n with sample data and mocked services; run your own test with a real meeting before using it with clients.

Want to start free? Download the WhatsApp agent and read 7 AI automations for small businesses.