How do I get a ChatGPT Dot to prep my meetings? A brief from calendar and mail before each call: 23 real cases
Name the call, point your ChatGPT Dot (OpenAI Dots) at the calendar event, docs and email threads, and have it list what it can reach before it writes. 23 real prep setups: scheduled sales briefs, last-minute phone prep, Meetings plugin notes and follow-ups after the call.

The short version: what a good meeting brief looks like, how to set it up, where it gets stuck
A ChatGPT Dot (OpenAI Dots) can read the calendar event, the mail, the files and last meeting's notes, and hand you a short brief before the call, then keep track of what was promised afterwards. This section sums up what people learned; the named cases follow below. To set up the Dot itself, start with getting started.
Is it worth setting up?
Yes, especially for external calls and for the follow-ups after them. The strongest cases have a fixed lead time and a narrow source list. チャエン gets a sales-prep report 30 minutes before every external meeting, with sources and an explicit "unknown"; Diana scored her Dot 9 out of 10 on a real planning meeting because it found everything she would otherwise have dug up; Shingo Irie gets talking points and the screens to show from last meeting's minutes. The weak spots are the shape of the output and anything that gets sent: briefs arrive as a wall of chat text unless you name the format, and a follow-up only becomes a draft you can approve when you ask for one. Let the Dot prepare and remind, and keep sends and calendar changes for your own yes.
How to set it up in seven steps
- Connect only what the brief needs: Google Calendar or Outlook for the event, Gmail or Outlook for the threads, Drive or Notion for the deck, and Slack if that is where the meeting was agreed (Dots + Google Calendar, Dots + Gmail).
- Add a custom rule so that sends, invites and calendar changes ask you first. Hrishikesh Bhat set permissions and approval rules before he scheduled a single brief.
- Run one brief by hand. Name the meeting, point the Dot at the calendar event, docs, threads and notes it may use, and ask it to list what it can and can't reach before it writes. OpenAI's Prepare meeting briefs use case gives the shape: objective, attendee context, source-backed facts, likely agenda, open questions, decisions you may owe and a notes template. The Meet dots guide uses the same shape for an offsite: read the plan and venue emails, list decisions, deadlines and what you're waiting on, and draft replies for review.
- Name the container in the same message: one page, cards, a PDF, or a draft you can approve. Diana's facts were right both times; the format changed only when she asked.
- Filter the calendar before you schedule. Skip internal and cancelled events, as チャエン does, so you get a few briefs worth reading rather than one per meeting.
- Pick the lead time by where you will read it: 30 minutes for a skim on the way, 15 for a last look before joining, two hours if you want time to prepare. Give the time zone and an end date, then ask the Dot to confirm what it saved.
- Feed it last time's record. Minutes, call transcripts, meeting notes and Meetings plugin pages are where the commitments and dates sit, and they make the next brief far better than the invite alone.
Where it gets stuck, and what worked
- The brief arrives as a wall of text. Diana asked for visual cards and a one-page overview after the fact, and the visual version was still being built when she checked back. Say the format in the first request, and once it is right, save it as a skill, as 友くん did with his morning infographic.
- The follow-up is chat text, not a draft. Diana's Dot invented nothing but she had to ask a second time for the message, and when she said send, it sent at once. Ask for the follow-up as a draft in your Gmail drafts and keep sending to yourself.
- The transcript isn't there yet. curt.kennedy's Dot said in advance it would keep checking for up to an hour if Google's transcript was still processing, and tell him if capture had failed. Ask for that fallback whenever the job depends on a file another tool produces.
- A calendar job stops halfway. Kengo Harada's Dot found times across three Outlook calendars, held six slots and posted in Slack, but cancelling the extra holds stopped when the tool stopped. End the request with "then list any holds you couldn't cancel", and check the calendar yourself.
- Files saved somewhere you can't see. 坂本匡史 was told his files were in "Library", found nothing there, and had them delivered to a Google Drive folder instead; later decisions reached the spreadsheet but not the earlier Word documents. Ask for a link you can open and a note of which file is the latest.
- It works through a list on its own, before you remember what was on it. Rohan Attravanam gave his Dot a list one evening and woke up to a follow-up text already sent to a prospect; he had forgotten half of what he asked for. Put "ask before you send anything to a person outside my company" in the custom rules before you hand over a list of follow-ups.
- Meetings plugin notes have gaps. Sabrina Ortiz of The Deep View found the summary and follow-ups on the mark, but the transcript has no speaker labels and the audio can't be replayed. If voices are missing, check both microphone and system audio permissions on the Mac.
- The brief guesses. Ask for a source next to each fact and the word "unknown" where it isn't sure. Both チャエン's reports and Diana's test were usable because the Dot had to show where each fact came from.
What it can do, and what it can't yet
It can:
- Send a short brief before each external meeting, with sources, at a lead time you choose (チャエン, David Yu's two-hour brief on the meeting prep workflow)
- Get you ready by phone or chat when a call is minutes away (Takuma Fukuda, Clay Lehman)
- Build talking points from last meeting's minutes and progress since then (Shingo Irie)
- Catch double-bookings from meeting notes or the calendar and fix them after asking (Rajat, 多田啓二)
- Turn a transcript into commitments, a follow-up draft, or a recurring meeting series with agenda and invites (Clay Lehman, Troy Drummond)
- Compare a spec with a meeting transcript and prepare an estimate, then send check points on the morning of the next meeting (坂本匡史)
It can't yet, or not reliably:
- Produce a send-ready follow-up draft without being asked for one (Diana)
- Finish every step of a multi-person calendar job without leaving holds behind (Kengo Harada)
- Keep earlier documents in step with decisions made later (坂本匡史)
- Label speakers in a Meetings plugin transcript or replay the audio (Sabrina Ortiz, OpenAI Help)
One message to start with
Help me prepare for [meeting] on [date, time, time zone]. First list which of these you can open: the calendar event, [docs], [Slack thread], the last email thread with [name], and the notes from our previous meeting. Then give me one page: why we're meeting, who is attending and their role, what we agreed last time, what is still open, three questions to ask, and anything I should check first. Put a source next to each fact and write "unknown" where you aren't sure. Don't send, invite or change anything. If this is right, I'll ask you to send it [30] minutes before each external meeting, skipping internal and cancelled ones.
Quick answers
What is a ChatGPT Dot meeting brief?
A short, source-backed prep pack your Dot builds before a named call—objective, who is in the room, facts from calendar/docs/mail, open questions, decisions you may owe, and a notes template—instead of a vague "summarize my meetings" dump. OpenAI's own Prepare meeting briefs use case is the clearest place to start.
How do I prep for one named call?
Name the meeting, point the Dot at the calendar event, docs, Slack threads and email it may use, and ask it to list what it can and can't reach before it writes anything. Try it first with "What's my next meeting with [name], and what was the last email thread?" Connect Google Calendar plus Gmail (and Drive if that is where the deck lives) first (Getting started with your dot).
Which apps should I connect?
Only what the brief needs: Calendar for the event, Gmail/Outlook for threads, Drive/Notion for decks, and Slack if that is where the meeting was agreed. Set a custom rule to Ask before taking action (or Hand off) for any send or calendar change.
What is the Meetings plugin?
A macOS ChatGPT desktop plugin that takes notes through your Mac's mic and system audio—no bot joins the call—then writes a personalised summary and suggested action items into ChatGPT Space. Audio is deleted once notes are ready; consent is on you. Calendar is optional for reminders (The Meetings plugin).
Is this the setup page?
No. This page collects real meeting-prep cases people posted about their own Dots (scheduled briefs, last-minute phone prep, follow-ups, records between meetings), with a summary of what they learned at the top. For the one-page brief template and more setups, use the meeting prep workflow. If Dots is missing in your account, start at Dots not showing.
Briefs on a schedule
1. A sales-prep report 30 minutes before every external meeting (チャエン, AI研究所)
In a YouTube Short posted 2 Oct, the Japanese creator チャエン (@masahirochaen) calls this the most useful thing his Dot does for him. It finds external sales meetings in Google Calendar, skips internal and cancelled ones, pulls the other side's details from connected email, Slack, CRM and public sources, and posts a prep report to the chat 30 minutes before the start. The report covers the company overview, the contact's recent updates, the existing contract and past exchanges, recent news, and what to confirm on the day. He asks for sources and tells it to say what it doesn't know rather than guess, then reads it on the way to the meeting.
His prompt from the video description, translated:
Check Google Calendar for external sales meetings and send a prep report to this chat 30 minutes before each one starts. Use connected email, Slack, CRM (such as Salesforce) and public information. Keep it short: company overview, the contact's recent updates, our contract and past exchanges, recent news, and points to confirm on the day. Add sources, and mark anything unknown instead of guessing. Skip internal meetings and cancelled events.
2. A brief 15 minutes before each meeting, still under review (Hrishikesh Bhat)
After a 24-hour test, Hrishikesh Bhat (2 Oct) set permissions and approval rules first, then assigned a brief 15 minutes before each meeting covering the people, the purpose and useful background. He says plainly that he is still checking what those briefs deliver.
3. An unprompted meeting flag and a one-page scratchpad (MindStudio)
MindStudio's Dots guide describes a week of always-on use with Calendar, Gmail and Slack connected. On a Monday the Dot flagged an upcoming meeting and a related email without being asked. Later it compiled a personal scratchpad, one agenda pulling emails, Slack threads and calendar context, so there was a single place to check before a packed day.
4. The right facts in the wrong container (Diana, AI Agents Library)
Diana tested her Dot, Sparky, on ten business tasks on 30 Sep and 1 Oct and wrote it up in her OpenAI Dots review. The first was preparing her for a real editorial planning meeting: participants, background, prior decisions, open issues and evidence gaps. She scored it 9 out of 10 because Sparky found everything she would otherwise have had to dig up. The weak point was format: it arrived as a wall of text in the chat. When she said she wanted visual cards and a one-page overview, Sparky said it would make that her default, but when she checked back later the visual version was still being built.
Last-minute prep, by phone or chat
5. A phone call 10 minutes before a sales meeting (Takuma Fukuda, EXTAGE)
Takuma Fukuda, who runs the Kyoto marketing firm EXTAGE, published his call with his Dot (7 Oct) almost unedited. Ten minutes before a sales meeting he phoned it and asked what he was supposed to propose. In about three minutes it laid out the proposal, how the company president's interests differed from the division manager's, what to confirm in the meeting and a few small-talk openers. He discloses that the client company, its people and the earlier minutes were made up for the demo; the call itself is real. His own rule, from the description: the Dot researches, organises and drafts, and he handles anything that gets sent, published or paid.
6. A call in 15 minutes and no prep done (Clay Lehman)
Clay Lehman, who works in property management and real estate, opens his week-with-a-Dot video (3 Oct) with this: a call was 15 minutes away and he wasn't ready, so he asked his Dot to find the appointment and get him ready. Afterwards it turned the call transcript into a list of commitments plus a follow-up email draft. He posted the same moment as a short reel on 6 Oct.
7. A whole conference, from travel to meetings (Elvis, @omarsar0)
Elvis (@omarsar0, 29 Sep) tested Dots for a few days before launch and used it to organise his entire DevDay, from travel to setting up sessions and meetings. He singles out the proactive updates and the note-taking tools as what made it useful.
8. The on-call channel, five minutes before going live (Tibo Sottiaux, OpenAI)
Tibo Sottiaux, who works at OpenAI, replied on 30 Sep that during the DevDay demo he had been watching the stream while reading through the on-call channel his Dot had just flagged to him: production had gone down five minutes before the demo.
Built from what was said last time
9. Talking points from last meeting's minutes and dev progress (Shingo Irie)
Indie developer Shingo Irie (@iritec_jp, 1 Oct) has his Dot prepare meetings from the previous meeting's minutes and the development progress since then. It returns what to talk about and which screens to show, with URLs. The same Dot regularly checks Slack, GitHub and email for requests he read but never answered, and explains the background and next step for each.
10. A follow-through nudge linked to the call transcript (Andrew Southworth)
On 30 Sep Andrew Southworth posted that his Dot noticed a calendar reminder to reduce ad spend on a campaign, cross-checked it against his meeting transcripts, and messaged him that morning to make sure he followed through, with a link to the transcript. He hadn't asked it to do this.
11. Meeting notes that exposed a double-booking (Rajat)
Rajat (@rajat_rg18, 2 Oct) wrote no prompts and set no routines; he only named his Dot Lex. Within 24 hours Lex read his meeting notes as they came in, spotted that a Friday catch-up he had agreed to clashed with an invite on his personal email, flagged it, took his input and fixed it.
12. Meeting notes as the Dot's context (Alexander Embiricos, OpenAI)
When the Meetings plugin shipped on 6 Oct, Alexander Embiricos of OpenAI wrote that he uses it "so my dot has the context it needs and stays on top of followups". How the plugin works is in the section below.
22. Monthly review meetings created from the call transcript (Troy Drummond / Lu)
Troy Drummond (@troydrum01) (13 Oct 2026) finished a meeting where the team discussed their recurring monthly review meetings. Immediately afterwards his Dot, Lu, checked the transcript and created the monthly meetings with a full agenda, the correct invites, and the right dates — without him rebuilding the series by hand.
Sales calls, interviews and follow-ups
13. Reviewing sales calls and chasing forgotten contacts (Pascal Dumais)
Entrepreneur Pascal Dumais shows his Dot, Théo, in a French-language video (2 Oct). The first two of his five uses are meeting work: analysing his sales calls, and following up with people he had forgotten to get back to.
14. Sorting the good questions for her interviews (Ashe Magalhaes)
After three days with her Dot, Ashe Magalhaes (3 Oct, "not sponsored") wrote that it had been filtering thoughtful questions for her interviews, alongside coordinating with her Airbnb host. She keeps offloading small thoughts during the day and finds it does something useful with them.
15. A prospect followed up overnight (Rohan Attravanam)
Rohan Attravanam (5 Oct) gave his Dot a list one evening. By morning it had sent a follow-up text to one prospect on its own and was partway through emailing images to another. He says he had forgotten half of what he asked for.
16. Post-meeting follow-up: careful, but not send-ready (Diana, AI Agents Library)
In the same review, Diana gave Sparky a real meeting record and asked for decisions, open questions, action items, owners, deadlines and the passage behind each, with explicit and proposed commitments kept apart, plus a short follow-up message. It invented nothing. It was still her lowest score, 7 out of 10: the follow-up arrived as more text in the conversation rather than a draft she could approve, and she had to ask for it separately. Once she told it to send, it sent straight away. Her takeaway is to draft follow-ups in ChatGPT and keep the Dot for the work around the meeting.
Between meetings: records, check-ins and the calendar
The cases above are about one meeting. These are about what happens around them: the record after the call, the notes from a standing check-in, and the calendar the Dot keeps watching in between.
17. A transcript check booked before the meeting started (curt.kennedy, OpenAI Community)
On 1 Oct curt.kennedy scheduled his Dot to work on a virtual meeting the next day, to see whether it could replace what Motion does for him: record, transcribe and pull out actions. The Dot's reply set out the plan. It would check after the meeting, with up to an hour of flexibility, fetch Google's transcript, read it and send him its own summary of decisions, action items, owners and deadlines. If the transcript was still processing it would keep checking, and if capture had failed it would say so. His part was to join as host in his browser and accept any recording prompt; Motion would run alongside so he could compare.
18. A graduation-day draft from old course material and the planning meeting's minutes (友くん, KAITEK)
In the 8 Oct morning live of the GPTs研究会 group, written up by AI氣道, 友くん (KAITEK, an AI support consultant and former physical therapist) showed one of two jobs where his Dot did real work. He had it read the materials from his past courses and match them against the minutes of the meeting where they discussed how the course's graduation ceremony should go. The result was a first draft of the materials for the ceremony. In the same session he said he still sends invoices himself and keeps purchases, important messages, money and passwords for a person to check.
19. Morning, midday and evening check-ins logged, then sent back as a morning infographic (友くん, KAITEK)
友くん didn't know what to give a secretary at first, so he asked his Dot to sort his work into categories and write a daily report at his morning, midday and evening check-ins. It started logging the state of each line of work into a spreadsheet, kept in Google Drive or Notion, where he keeps everything. The Dot then told him he could save the request as a skill. He did, and asked for the report to arrive at a set time as an infographic grouped by category. The one he showed on the live told him what to do that day and listed what the Dot would do and summarise itself. For a client in healthcare, he said, it also proposes the next move.
20. "There's an overlap here, is that OK?" (多田啓二, AIMUNIQ)
On the same live, 多田啓二 (ただっち, AIMUNIQ) said his Dot, with Google Calendar connected, asks him whether he has forgotten something and points out double-bookings before they happen. He also said he is still getting used to working with it. Two more calendar clashes, caught from meeting notes and from Slack, are in case 11 above and on the meeting prep workflow.
23. Three people's Outlook calendars, six holds, one fixed slot and an attendance check in Slack (原田 憲悟 / 自認カービィ)
Kengo Harada (原田 憲悟, @kenboo) wrote on note (13 Oct, Japanese) about a scheduling job he calls a real chore. His Dot, 自認カービィ, checked the Outlook calendars of the three people involved, found times when everyone could attend, put tentative holds on six candidate slots and sent the invites. He then told it to fix one slot, cancel the remaining holds and ask the attendees in Slack to confirm they could come. Moving to the fixed slot and the Slack post both went through. Cancelling the other holds stopped partway when the tool stopped, so he is careful to say the whole job did not finish. Even so, he says, finding candidate times and sending provisional invites already takes most of the work off a person. If you ask for the same thing, end the request with "then list any holds you couldn't cancel", and look at the calendar yourself afterwards.
After the meeting: from the spec and transcript to an estimate
21. A spec, a transcript and an estimate, then check points on the morning of the next meeting (坂本匡史)
坂本匡史 (Masashi Sakamoto), an AI solopreneur in Fukuoka who builds systems for clients, began with a Dot that proposed three jobs it could take on from email and meeting notes. Sakamoto picked one (note, 4 Oct, in Japanese): get the estimate ready for a development project discussed the day before. The Dot compared the specification PDF attached to an email with the Google Meet transcript of the meeting, pointed out that three options in the spec had become two in the meeting, and logged the hearing results, the meeting history and the next action in the sales-management system Sakamoto already uses. Of 13 hearing items, 8 were confirmed and 5 were still open, so it left the checklist as a draft. They then worked through the scope together: how much source code and handover material to provide, how much explanation to include, and how far maintenance goes. Sakamoto made each call, the Dot wrote it into the documents, and the result was an estimate-prep document, a maintenance proposal, an in-scope and out-of-scope list, estimates for both options, a table of scope by development phase, and an AWS cost estimate made by operating the official pricing calculator. Sakamoto spent about four hours on it across two days, against roughly a week the job would normally take by Sakamoto's own reckoning (not a measured comparison), and nothing had gone to the client yet. Two snags: the Dot said the files were saved to "Library", but there was no such entry on Sakamoto's screen, so they were delivered to a Google Drive folder instead; and conditions decided later reached the spreadsheet but not the Word documents written earlier.
In a post from 7 October, the Dot sent check points on the morning of another meeting. From an earlier investigation report: image display had been restored, but the change to stop the problem recurring hadn't been made, and display coming back didn't prove the update process was working, so the write side was worth checking. For an estimate due soon on a different project, it reminded Sakamoto to submit a copy without the internal costing sheet.
Meetings plugin: notes your Dot can work from
OpenAI added a Meetings plugin on DevDay (release notes) and @ChatGPT announced it on 6 Oct. What the help article says:
- It runs in the ChatGPT desktop app on macOS, in beta for all Pro and Business plans. Enterprise is in a limited alpha; iOS, Android and Windows are listed as coming soon.
- No bot joins the call. It listens through your Mac's microphone and system audio, so it works for online calls and in-person conversations.
- When you stop, it writes a personalised summary and suggested action items, such as a follow-up email draft or a project-plan update, onto a page in ChatGPT Space next to your own notes. Pages are private until you share them, and people you share with see the notes but not the transcript.
- The audio is deleted once the notes are ready. Telling participants and getting their consent is up to you; the in-app reminder is only shown to you.
- A calendar is optional. Connect the Google Calendar or Outlook Calendar plugin if you want reminders and a list of upcoming meetings.
- To set it up, open Plugins in the desktop app's sidebar, install Meetings, go through onboarding and allow both microphone and system audio. If voices are missing from a transcript, one of those two permissions is usually off. On Business plans, ChatGPT Space and the Meetings plugin both have to be enabled for the workspace.
- It can notice that you're in a call and offer to take notes even when the meeting isn't on your calendar. It stops on its own when you close the lid, after a long silence, or when a session reaches four hours, and the notes appear a few minutes after the audio has uploaded.
- On the meeting page you can edit each suggested action item and start it with one click, or ask ChatGPT questions such as when you agreed to schedule a follow-up.
Sabrina Ortiz of The Deep View tried it on a quick catch-up call with a colleague (7 Oct). The summary and follow-up items were on the mark and the transcript was faithful word for word, but it has no labels for who was speaking, no way to add them, and the audio can't be replayed. She treats it as a stand-in for Granola rather than Otter.ai: good for a rundown and for asking what was said, not for an audio record.
More asks you can copy
Replace the brackets before you paste.
A brief before external meetings:
Every weekday, find my external meetings in Google Calendar for the next day, skipping internal and cancelled ones. Thirty minutes before each, send me a one-page brief: who is attending and their role, what we agreed last time (from email, Slack and meeting notes), what is still open, three questions to ask, and anything I should check first. Put a source next to each fact and write "unknown" where you aren't sure. Time zone: [zone]. Run this until [date].
Last-minute prep by phone:
My call with [name or company] starts in 15 minutes. Find the calendar event and the last three email threads with them, then tell me in two minutes: why we're meeting, what they care about, what I owe them, and one thing to confirm before we hang up.
After the meeting:
Read the notes from today's [meeting name]. List decisions, open questions and action items with owner and date, and quote the line each one comes from. Keep commitments people actually made separate from ones you're suggesting. Then put a short follow-up email in my Gmail drafts for me to review.
A transcript check booked before the call:
After my [meeting name] on [date, time, time zone] ends, check for the [Google Meet / Zoom] transcript within the next hour. If it's still processing, keep checking until [time]; if there is no recording, tell me straight away. When you have it, send me decisions, action items with owner and deadline, and the open questions, each with the line it comes from.
A daily check-in log that becomes a morning summary:
Sort my work into [three to five categories]. At [morning, midday and evening times, time zone], ask me for a one-line update on each and log it in [sheet or Notion page] with the date. Every morning at [time], send me one summary grouped by category: what I should do today, what you will do yourself, and anything that has been stuck for more than [two] days. Once the format is right, offer to save it as a skill.
Related
- Meeting prep workflow: a one-page brief template plus more setups, including meeting docs ready on waking, scheduling rules, a promise caught in a class transcript and clashes caught from Slack
- Dots + Google Calendar: setup and real cases, including two executive interviews that clashed and got new prep briefs, and a weekend "zombie meeting" the Dot offered to rebuild
- AI secretary & daily catch-up · Call your Dot · Email triage · Small business & clients: real cases · Project follow-up
- Work trips and travel days: real cases, including a Dot that prepped every meeting during a 12-hour flight
- Use cases hub · Getting started · Pricing
Copy this task brief
How to get a ChatGPT Dot meeting brief from calendar and mail, where it usually gets stuck, how the Meetings plugin fits, and real prep cases: scheduled briefs, phone prep, follow-ups and records between meetings.Download Markdown ↓