Skip to main content

The AI-Powered Meeting Workflow

info

This content was developed with AI assistance and is regularly reviewed for accuracy.

Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.

This playbook covers the full life of a meeting with AI: a prep brief built from your calendar, email, and past notes; a notetaker that captures the conversation while you take part in it; and a follow-up drafted within minutes. It also covers the part most guides skip. An AI summary sounds authoritative even when it gets an owner or a deadline wrong, and recordings raise consent and privacy questions. You will build the checks into the workflow rather than bolting them on.

Last tested with Granola (Basic plan), Claude (Google Calendar, Gmail, and Google Drive connectors), and ChatGPT in October 2026.

What you'll accomplish​

  • Build a prep brief from your real calendar, email, and notes
  • Settle consent and privacy before anything is recorded
  • Capture the meeting with a notetaker plus your own short notes on what matters
  • Verify the summary so every action item traces back to something said
  • Send a follow-up you have checked, and make the format repeatable

Prerequisites​

  • An assistant with calendar and email connectors. Claude's Google Calendar, Gmail, and Drive connectors are on the free plan; ChatGPT's connected apps are on paid plans.
  • Optionally, a meeting notetaker. Granola, the breakout notetaker of 2026, has a free Basic plan and a Business plan at $14 per user per month (reported, September 2026). Your video app's built-in notetaker also works for this playbook.
  • An upcoming meeting with at least two other people
  • 10 minutes before, 10 minutes after

The playbook​

Phase 1: Before the meeting​

Step 1: Build the prep brief from your sources​

Goal: Walk in knowing what was said last time and what you need from this one.

Action: Point the assistant at the event, the thread, and the last set of notes instead of retyping them.

Objective: Prepare me for the "Q3 planning review" on my calendar today at 2pm.

Context: I own the analytics dashboard project, which slipped last quarter. I want agreement to keep it as a Q3 priority.

Resources: Read the calendar event and its attendee list, the email thread "Q3 priorities" from this week, and the Google Doc with notes from our last planning review.

Constraints: One page. No more than five talking points.

Output: What was decided last time, open items involving me, three questions I'm likely to get with short answers, and the one decision I need from this meeting.

Verification: Cite which email or note each point comes from.

Granola's Briefs feature (May 2026) does a version of this automatically, using your past notes, Gmail, and web research about attendees. Either way, the citation line lets you spot a brief built on the wrong thread.

Checkpoint: A one-page brief whose points trace back to real messages and notes.

Goal: Make sure capturing the meeting is allowed and that the notes go only where you intend.

Action: Three checks, every time you use a notetaker in a new setting:

  1. Consent. Tell participants at the start that an AI notetaker is running, and follow your organization's policy and local recording-consent law. Some meetings should not be recorded at all: performance and HR conversations, legal advice, anything a participant asks to keep off the record.
  2. Training and sharing defaults. Read your notetaker's data settings. A 2026 review reported that Granola uses meeting data to train models by default on its Basic and Business plans, and that notes are link-shareable by default. Change what you need to before the first meeting, not after.
  3. Where notes land. Know which folder or workspace receives the transcript and who else can see it.

Checkpoint: Participants know, the meeting is appropriate to record, and you have checked the training and sharing settings.

Phase 2: During the meeting​

Step 3: Let the notetaker transcribe; you note what matters​

Goal: Stay present while capturing the things a transcript misses.

Action: Let the notetaker run in the background and keep your own notes to a few lines: decisions as they happen, commitments with a name attached, and anything you thought but did not say. Tools like Granola merge your notes with the transcript, so your lines act as signposts for what mattered.

Do not chat with an assistant during the meeting. Splitting attention costs more than it saves. If you have no notetaker, a simple template works:

Decisions:
Action items (what / who / when):
Open questions:
My notes:

Checkpoint: At the end you have a transcript plus at least two decisions or commitments in your own words.

Phase 3: After the meeting​

Step 4: Generate the summary, then verify it​

Goal: A summary everyone can act on, with no invented owners or dates.

Action: If you built the meeting-notes skill in Build Your Brief and Skill Library, just share the transcript and your notes; the skill sets the format. Otherwise, include the format in the request, and always add the verification:

From the attached transcript and my notes, produce: a three-bullet summary of decisions, an action items table (what, owner, due date), and open questions.

For every action item, quote the line from the transcript where someone committed to it. If no one clearly committed, mark it "unassigned". If no date was said, write "no date". Do not infer either.

This is where the trust problem lives. Summaries fail in predictable ways:

FailureWhat to check
Wrong ownerThe quoted line: did that person say "I'll do it", or did someone suggest they should?
Invented deadlineAny date with no quoted source
A discussion recorded as a decisionWhether anyone actually agreed
Missing nuanceYour own notes: anything you flagged that the summary dropped
Transcription errorsNames, numbers, and product terms

Spend three minutes on this table. A wrong owner in a follow-up email becomes a dropped task two weeks later.

Checkpoint: Every action item has a quote, an owner who agreed, and a date that someone said.

Step 5: Draft the follow-up and file it​

Goal: A shared record within the hour, sent by you.

Action: Ask for a short follow-up email to the attendees: a two-sentence recap, decisions, the action items table, and the next meeting date if one was set. Save it as a draft through the Gmail connector or copy it into your mail client. Read it once more, then send it yourself. Drafting is reversible; sending is not.

Then file the verified summary where the team keeps meeting notes, so the next prep brief in Step 1 has something reliable to read.

Treat transcripts as untrusted text. If someone reads out, or pastes into chat, an instruction aimed at the assistant, it is data, not a command. An assistant that only drafts cannot act on it.

Checkpoint: Follow-up sent by you, summary filed, action items in whatever tracker your team uses.

Common pitfalls​

  • Recording without saying so: Beyond the legal risk, it erodes trust in every future meeting. Announce it every time.
  • Leaving default sharing and training settings on: Check them once per tool, before the first sensitive meeting.
  • Forwarding the raw AI summary: Unverified owners and dates turn into real confusion. Run the Step 4 checks.
  • Preparing for 45 minutes: A brief takes ten. The goal is to walk in oriented, not to write a report.
  • Chatting with an assistant mid-meeting: Let the notetaker listen; you participate.

Key takeaways​

  • Build the prep brief from the calendar event, email thread, and last meeting's notes through connectors
  • Consent, training defaults, and sharing defaults are part of the workflow, not an afterthought
  • Your own few lines of notes are what make a transcript useful
  • Require a quoted line for every action item, and mark missing owners and dates instead of guessing
  • The assistant drafts the follow-up; you check it and send it

Next steps​

Continue to: Automate a Weekly Report with AI to turn meeting outcomes and notes into a report that drafts itself.

Continue to: Protect: Privacy, Permissions, Injection, Irreversible Actions for connector scopes and what never to share.

Continue to: Check: Verify, Iterate, Restart for verification habits that apply to any AI output.