What a dot actually is
Regular ChatGPT waits for you to type. A dot does not. It runs on GPT-6 Astra, has its own cloud computer and browser, connects to the apps you pick (over 4,000 are supported through ChatGPT's plugins), and keeps working between your conversations.
When you are not talking to it, it does what OpenAI calls proactive research: it reads your connected apps looking for ways to help. When it finds something, it messages you. You approve, and it does the work.
That is the whole shift. You stop being the one who has to remember everything.
Can you get it right now?
- Pro: Yes, except in the EU (EEA), Switzerland and the UK.
- Business Premium: Yes, in every region ChatGPT supports.
- Enterprise, Edu, Healthcare: Beta, but it starts turned off. Your workspace admin has to switch it on.
- Plus and Free: Not yet. OpenAI says more plans are coming.
Your first dot is included in the plan at no extra cost. You have to be 18 or older.
1. The 7-step setup
About 20 minutes, done once. Steps 4 and 5 are where most people will cut corners. Don't.
- Create it on your computer, not your phone
Open the ChatGPT desktop app (Mac or Windows) or go to chatgpt.com/dots in a desktop browser. You cannot create a dot on mobile, and it does not work on mobile web. Your phone only works after the first setup. - Name it
You pick a name and a character during onboarding. Its handle starts as @yourname-dot and becomes @yourname-whatever-you-named-it. You can change the name or avatar later from its profile with the pencil icon. - Connect only the apps you want it reading
Go to the Plugins tab. These are the same connections ChatGPT, ChatGPT Work and Codex use, so anything you already connected is already there. Be picky: your dot reads connected apps in the background and forms memories from them. Start with email, calendar and Slack. Add more once you trust it. - Brief it with Prompt 1 below
It already pulls in what ChatGPT remembers about you. The briefing fills in the rest: your goals, your standards, and when it should (and should not) bug you. This is the step that separates a dot that helps from a dot that spams you. - Set your Custom Rules with Prompt 2 below
Rules decide what it can do alone and what needs your OK. On mobile it lives under Customize, then Custom rules. Your dot can draft the rules for you, but it needs your approval to save them. - Hook up the places you actually talk
From the desktop app, connect Slack (Teams is supported too). After that you can message it or voice call it from the ChatGPT mobile app. Texting is a limited beta for Pro users in the US only, so do not count on it yet. - Optional: let it use your own computer
This starts turned off. Turn it on from the desktop app on that computer and confirm Allow access. Then it can work with files on your desktop and use your local browser. Revoke access anytime. Skip this on day one.
2. Prompt 1: The briefing
A dot is only as good as what it knows about you. OpenAI's whole pitch is that it does work "the way you would do it." It cannot do that if it has never heard how you work. Paste this as your very first message and answer honestly. Real names, real numbers, real examples.
Before you take on any work for me, I want you to actually know me. Interview me one question at a time and wait for my answer before asking the next one. If an answer is vague, push back and ask for a real example, a real name, or a real number. Cover these, in this order: 1. What I do, who I do it for, and what a great week looks like for me. 2. The 3 projects or goals that matter most right now, and what "done" looks like for each one. 3. The people who matter most (my team, clients, family) and how I like to communicate with each of them. 4. My standards: how I write, what I hate seeing in a draft, and what "good" looks like to me. Ask me to point you to 2 or 3 things I've written that I'm proud of. 5. The recurring stuff that falls through the cracks for me (invoices, follow-ups, renewals, birthdays, bills). 6. Anything you should never do without asking me, and anything you should never touch at all. 7. When to reach out: what's worth interrupting me for right away, what can wait for a daily summary, and my quiet hours. When we're done, write it all up as "How I Work" in one page and show it to me for edits. Then look through the apps I've connected and give me the 5 responsibilities you think you could take off my plate first, ranked by how much time they'd save me. For each one, tell me exactly what you'd do, how often, and what you'd need my approval for. Don't start any of them until I pick.
3. Prompt 2: Your Custom Rules
Your dot comes with default rules. Custom Rules let you tighten or loosen them. Every rule gets one of four behaviors:
Take action without asking. Take action if pre-approved (meaning you asked for that exact thing in your message). Ask before taking action. Hand off to you.
Start strict and loosen up as it earns your trust. Here is the starter pack I would use:
Take action without asking
- Read and summarize my email, calendar, Slack and connected docs
- Draft replies and documents and show them to me in our chat
- Create files and do research on your own cloud computer
Take action if pre-approved
- Send a message I specifically asked for, to the person I named
- Run reminders and recurring checks I set up
Ask before taking action
- Send any email or Slack message to anyone who is not me
- Accept, decline or move a calendar event
- Buy anything, even with a card I saved on that site
- Share a file or change a shared doc
- Delete anything
Hand off to you
- Signing contracts or agreeing to terms on my behalf
- Anything involving my bank or moving money
Or skip the typing. Your dot can write the rules for you based on your briefing and your connected apps. It still needs your approval to save them.
Help me set up my Custom Rules. Use my "How I Work" notes and look at every app I've connected. Propose a list of rules sorted into the four behaviors: Take action without asking, Take action if pre-approved, Ask before taking action, and Hand off to me. Start strict. If you're not sure where something goes, put it under "Ask before taking action." Make sure these are covered: sending any message to anyone other than me, calendar changes, purchases, sharing files, editing shared docs, and deleting anything. Write each rule as one plain sentence and tell me in one line why it belongs in that bucket. Then walk me through them one at a time. Don't save any rule until I approve it.
4. 10 jobs to hand your dot
Every one of these comes from something OpenAI showed or a real early tester did. Fill in the brackets, paste it to your dot, and it takes it from there. Every prompt that sends, buys or changes something stops and checks with you first.
1. The morning brief that catches conflicts
Every morning it tells you what is coming up and flags the collisions before they bite: two meetings on top of each other, a work call during something personal, a deadline nobody prepped for.
OpenAI's own demo: a dot notices a finance review now overlaps a daughter's recital, finds a time that works for everyone, and moves it once approved. Dan Shipper's dot at Every caught a calendar conflict with a video shoot from a Slack discussion.
Needs: Connect your calendar, email and Slack
Every weekday at 7:30am, send me a morning brief. Keep it under 10 lines. 1. Today's calendar, in order, with anything I need to prep for marked. 2. Conflicts: anything overlapping, anything scheduled on top of a personal event, and any meeting where the time or place changed in email or Slack but not on my calendar. 3. The 3 emails or Slack messages that actually need me today, with one line on why each one matters. 4. Anything due in the next 48 hours that I haven't touched. For every conflict, suggest a fix (a new time that works for everyone, or who to ask). Don't move, accept, or decline anything until I say yes.
2. The money you forgot to ask for
It watches for work you finished but never billed, invoices nobody paid, and payments that are late. Then it drafts the invoice or the nudge for you to approve.
This is the real launch story: an early tester's dot noticed he forgot to invoice a publication, prepared the invoice, and sent it after he approved.
Needs: Connect email, plus wherever you invoice from if it's in your Plugins list
Own my invoicing. Once a week, and any time you notice something in the background: 1. Look through my email and connected apps for work I delivered (finished projects, published pieces, completed calls) that I haven't invoiced yet. 2. Look for invoices I sent that are past due. For anything unbilled, prepare the invoice using my past invoices as the template (same format, same rates unless I've agreed to something different in writing). For anything overdue, draft a short, friendly follow-up in my voice. Show me each one with the amount and who it's going to. Never send an invoice or a payment reminder until I approve that specific one.
3. Inbox and Slack triage
Instead of 40 unread things, you get the handful that need you, plus drafted replies. Everything else becomes one line.
In OpenAI's Teams demo, the dot turns six unread DMs into the three that actually need attention and drafts the update the manager asked for.
Needs: Connect email and Slack (or Teams)
Twice a day, at noon and 5pm, go through my unread email and Slack DMs. Sort everything into: - NEEDS ME: someone is waiting on a decision, an answer, or something only I can do. Tell me who, what they need, and by when. - DRAFTED: things you can answer from what you already know about my work. Write the reply in my voice and show it to me. - FYI: one line each. - IGNORE: newsletters, promos, automated notifications. Just give me the count. If something is urgent (a client is upset, a deadline moved, money is involved), message me right away instead of waiting. Never send a reply without my OK.
4. Turn one recording into a week of content
Drop a transcript or recording and it finds the moments worth clipping, writes the show notes and drafts the social posts. It carries your edits forward so the next batch needs fewer.
This is the content-creator example straight from OpenAI's launch post and feature page.
Needs: Attach the transcript with + in the conversation, or point it at a connected Drive folder
Here's a transcript from [my latest podcast episode / video / interview]. I make content for [who my audience is, e.g. small business owners who are new to AI]. 1. Find the 5 best moments to clip. For each one, give me the timestamp, a hook line for the first 3 seconds, and one sentence on why it'll land with my audience. 2. Write show notes: a 2-sentence summary, 5 key takeaways, and any links or tools mentioned. 3. Draft 3 social posts in my voice. Match the tone of my past posts if you can see them. Keep everything as drafts for me to approve. When I edit something, remember what I changed and apply it to the next batch without me asking.
5. Catch the thing that no longer matches
Plans change and the emails, docs and decks do not. It notices when something you are about to send still says the old thing.
OpenAI's example: before a launch meeting, the dot notices the customer email still promises a feature that was cut, checks the latest decisions, and prepares the fix.
Needs: Connect Slack, email and wherever your docs live
I'm working on [the project, e.g. our November product launch]. The source of truth is [where decisions live, e.g. the #launch Slack channel and the Launch Plan doc]. Whenever a decision changes (scope, dates, pricing, who owns what), check every related draft: emails, docs, decks, landing page copy. If anything still says the old thing, tell me what's wrong, where it is, and show me the fixed version. Don't edit shared docs directly. Bring me the changes to approve first.
6. Keep your numbers deck current
The weekly report, the investor update, the leadership deck. It refreshes the numbers as they land and flags what changed and why, so you are never presenting last month.
From OpenAI's feature page: sales are up but fewer customers are coming back, and the dot checks the numbers against customer feedback and updates the leadership deck.
Needs: Connect wherever the numbers and the deck live
Keep my [weekly report / investor update / leadership deck] current. The numbers come from [the source, e.g. our Stripe dashboard and the Metrics sheet]. Every [Monday morning]: 1. Update every number in the deck to the latest data. 2. Flag anything that moved more than [10]% from last period. 3. For each big move, dig into what might explain it (customer feedback, support tickets, what changed that week) and add a short "what might be going on" note with what we still don't know. Show me the updated version before you share it with anyone.
7. Follow a story and draft the write-up
Pick a topic that keeps moving, like your industry, a competitor, or a law that affects you. It keeps up with it and hands you a draft when something actually changes.
Following a developing story and preparing a newsletter draft is one of the examples in OpenAI's safety post on dots.
Needs: Nothing extra. It uses its own browser
Keep up with [the topic, e.g. new AI rules for small businesses in the US / what my top 3 competitors are shipping]. Check a few times a week. Only message me when something actually changes, not just when there's new coverage of the same thing. When something does change: 1. Tell me what happened in 3 sentences, with links to the original source. 2. Tell me why it matters for [me / my business / my audience]. 3. Draft a short write-up I could send to [my newsletter / my team], in my voice. If nothing meaningful happened this week, send me one line on Friday saying so.
8. Watch your travel plans
Flight changes, gate changes, a hotel confirmation that never showed up. It spots the problem in your inbox before you are standing at the airport.
OpenAI uses "noticing a change to your travel plans" as its main example of what dots do in the background. Dan Shipper also had his dot help book a flight.
Needs: Connect email and calendar
Keep an eye on any trip I have coming up. Watch my email for flight, hotel, and rental car confirmations and any changes to them. Message me right away if: - A flight time, gate, or seat changes, or a flight gets canceled - Something I booked is missing a confirmation - A reservation doesn't match my calendar (wrong dates, wrong city) The day before I leave, send me a one-screen summary: flight times, confirmation numbers, hotel address, and anything I still need to do. If a rebooking is needed, find the options but don't book or pay for anything until I approve.
9. Turn customer requests into tested fixes
For anyone with an app or a site: it watches for the requests people keep making, scopes the small ones, builds and tests them, and brings you finished changes to review.
This is OpenAI's developer example: complete pull requests with videos showing the change, while you keep working on the big stuff.
Needs: Create a Codex cloud environment for your project first, then connect your feedback source
Watch [where feedback comes in, e.g. our #feedback Slack channel and support inbox] for requests and bug reports about [my app]. Once a week, group what you found by how many people asked for it. For the top requests that are small (a fix or a tweak, not a new feature): 1. Scope it in 3 bullets. 2. Build it in the [project name] Codex environment and test it. 3. Bring me the pull request with a short video or screenshots showing the change, plus any trade-offs. Don't merge or ship anything. Anything bigger than a small fix, just add to a list for me.
10. Dinner when you are working late
The fun one. It pulls your usual orders with the full total and delivery time, and places the order once you pick.
Straight from OpenAI's "day with your dot" demo.
Needs: DoorDash connected, if it shows up in your Plugins list. Every purchase needs your approval
If I'm still working after 7pm on a weekday, check in and ask if I want dinner. Show me my 3 usual orders from DoorDash with the full total (fees and tip included) and delivery time for each. Place the order only after I pick one and say yes.
5. Prompt 3: The Friday tune-up
OpenAI says your dot learns from feedback over time. The catch: you cannot open its memory and edit it. The only way to teach it is to tell it. So make that a 5-minute habit once a week, and it gets noticeably better every week you do it.
Set this up as a recurring task every Friday at 3pm. Look back at everything you did for me this week, including your background research, and tell me: 1. The 3 things you did that saved me the most time. 2. Anything you flagged that turned out not to matter, so you can stop flagging that kind of thing. 3. Anything you got wrong or weren't sure about. 4. One new responsibility you think you could own next week, based on what you saw in my apps. Then ask me for feedback on each one, one at a time. Update how you work based on my answers and tell me in one line what you're changing.
9 things that trip people up
- You cannot set it up on your phone Create it in the ChatGPT desktop app or in a desktop browser. Mobile works after that. Mobile web does not work at all.
- Do not see it yet? You might just be early OpenAI says access can take several days to reach every account. And if you are on Pro in the EU, Switzerland or the UK, it is not available to you yet. Business Premium works everywhere.
- Connecting an app is a bigger deal than it looks Your dot reads connected apps in the background and forms memories from them, even when you have not asked about them. Disconnecting an app stops new reading, but it does not erase what it already learned. Only a full Reset does that.
- You cannot see or edit its memories one by one If it gets something wrong about you, tell it directly and ask it to correct how it works. The only way to wipe its memory is Reset, which deletes the dot, its conversations and its scheduled tasks. It also shares memory with regular ChatGPT. Turning off Memory in ChatGPT stops the sharing but does not delete what it already has.
- Approving one message is not a blank check Saying yes to one email does not give it ongoing permission to contact people. If you want to approve something ahead of time, be specific: who it goes to, what it says, and when or under what conditions it sends.
- Some things always come back to you Changing a password and moving money are handed back to you every time. No Custom Rule can change that. Deleting data and installing software ask for approval each time. That is on purpose.
- Background work is read-only When it researches on its own, it can only read. It cannot send messages, edit anything, or drive a browser in the background. Anything it wants to actually do goes through your rules first. So if it finds something, expect a message, not an action.
- It cannot call you You can voice call your dot, but at launch it cannot call you. It reaches you with a message in ChatGPT or Slack, so turn on ChatGPT notifications on your phone or you will miss the whole point.
- Chatting is free, heavy work is not Talking to your dot does not count toward your ChatGPT limits. Tasks it kicks off in Codex or ChatGPT Work do. For the first month after launch, OpenAI is not counting dot usage against Pro and Business plans at all, so this is the month to experiment.
Where this came from
The use cases come from OpenAI's launch post, the ChatGPT dots page, and OpenAI's safety post on dots. The calendar and flight examples come from Dan Shipper's Vibe Check at Every.
The setup steps, rules and gotchas come from OpenAI's help articles, Getting started with your dot and Dots privacy, security, and safety FAQs, current as of September 29, 2026. Dots launched that day and OpenAI is changing things fast, so if a menu has moved, the help center will have the latest.