Your dot, properly understood

1. Step 1 of 11

   A chat answers; a dot keeps going

   A normal ChatGPT chat answers and then waits for you. A dot takes ongoing responsibility:

   • It keeps making progress between conversations.
   • It decides when to pause and when to wake up and continue.
   • It contacts you when it needs a decision.

   The kind of job that suits a dot is "keep an eye on it":

   • a trip's deadlines
   • replies about an event
   • a morning briefing

   A quick one-off question is still better in a normal chat.

   Think of it like…

   A chat is a reference librarian you visit with a question. A dot is an assistant on staff who keeps your projects moving while you are away.

   You’re done with this step when

   You have one ongoing responsibility in mind that you would like to hand off.

   I did this

2. Step 2 of 11

   Create your dot

   1. On a computer, sign in to ChatGPT in the desktop app or a desktop browser.
   2. Open Dots and follow the introduction.
   3. Your dot gets a name and a handle like @yourname-dot. You can rename it and change its look from its profile.

   Talking with your dot does not count against your normal ChatGPT limits.

   You’re done with this step when

   Your dot exists and has introduced itself.

   I did this

3. Step 3 of 11

   Write a good standing instruction

   OpenAI's advice: start with ongoing work, not one-off requests, share the sources it needs, and say which decisions are yours.

   A good instruction covers:

   • scope: what to watch or do
   • how often
   • for how long
   • the line it must not cross without you

   For anything that repeats, always give a time, a timezone and an end date.

   Think of it like…

   A job description, not an errand: what you own, how often, until when, and what to check with me.

   See an example: Three good standing instructions ▸See an example: Rewriting a vague instruction ▸

   Try this

   For the next four weeks, every weekday at 9:00 AM London time, check my email for anything about the conference, summarise what needs me today, and draft replies but do not send them.

   Copy

   You’re done with this step when

   Your dot accepted the job and told you when it will first report back.

   I did this

1. Step 4 of 11

   Its own computer in the cloud

   Your dot has its own computer and browser in the cloud. That is why it keeps working when your laptop is shut. It keeps:

   • its files
   • installed programs
   • its browser sign-ins between jobs

   You can watch it work, or take over the mouse and keyboard and hand control back.

   Being logged in somewhere in your browser does not log the dot in. When a site needs a login, the dot asks you to type it into a private form, so your password never appears in the chat.

   Think of it like…

   A desk in the office that is always staffed, even after you have gone home.

   Try this

   Open your dot’s profile, go to Computers, choose Open computer, then Take over and Return control.

   Copy

   You’re done with this step when

   You have watched your dot’s computer, and you know how to take over and give it back.

   For the curious

   You can also let the dot use your own computer: Computers, then Allow access, and Revoke access to remove it.

   Only one personal computer can be connected at a time, and it works only while that computer is online with ChatGPT open. Connecting it does not give the dot your past conversations.

   I did this

2. Step 5 of 11

   Three kinds of work: its own, handed off, and scheduled

   A dot juggles three kinds of work:

   • Assigned work: the ongoing responsibilities you gave it. It may split these across helpers running in parallel.
   • Delegated tasks: separate jobs it hands off, for example to Codex or Work, or onto your computer. A delegated task gets its own written brief and does not see your whole conversation.
   • Scheduled tasks: saved to run at a set time, or when something happens, like a message arriving in Slack.

   OpenAI's own warning: a task marked done does not prove it was delivered. Check the result.

   Think of it like…

   Your own to-do list, jobs you hand a contractor with a written brief, and a recurring calendar reminder.

   You’re done with this step when

   You can say which of the three your morning briefing is.

   For the curious

   Delegated jobs in Codex or Work count against those products’ limits, even though chatting with the dot does not.

   I did this

3. Step 6 of 11

   How it remembers: conversation, ChatGPT memory, and its notes

   A dot draws on three layers:

   1. The current conversation.
   2. Your ChatGPT memory: what ChatGPT already knows about you.
   3. Its own dot notes: your preferences, decisions and ongoing projects. These are kept separately from ChatGPT's memory and updated as things change.

   The notes follow the dot everywhere. Talking to it in Slack instead of ChatGPT does not reset them, though the visible conversations stay separate.

   Think of it like…

   ChatGPT memory is your general profile on file. The dot’s notes are its own working notebook for your projects.

   Try this

   What have you noted about how I like my summaries written? Correct anything that is out of date.

   Copy

   You’re done with this step when

   Your dot told you what it has noted, and you corrected one thing.

   I did this

4. Step 7 of 11

   What it can see at once, and why to correct it

   Like every AI model, a dot works from its context: the messages, instructions, sources and tool results in front of it at that moment. It picks out what seems relevant, which can be less than everything that exists.

   OpenAI does not publish how much a dot can hold. So treat its notes as a working summary that can drift, and review them now and then, especially after a long project.

   Think of it like…

   The papers on the desk, not the whole filing cabinet.

   You’re done with this step when

   You know why a dot can miss something, and you have a habit for catching it.

   I did this

1. Step 8 of 11

   Apps and permissions: give it only the keys it needs

   A dot can use the apps you have connected in ChatGPT, such as email, calendar and files, but only within those apps' permissions.

   • Let it read email without letting it send.
   • Each connection is separate. Adding Slack does not give it your inbox.

   Connect one app for your first job, not everything.

   Think of it like…

   Giving a house-sitter the mailbox key, but not the car keys.

   You’re done with this step when

   You connected one app your first job needs, with the least access that works.

   I did this

2. Step 9 of 11

   Custom rules: what needs your OK

   Before an action that could touch your accounts or share information, a built-in review decides: let it happen, ask you, or hand it to you.

   Changing a password or moving money always comes to you.

   You can add your own rules in Settings > Personalization > Custom rules, at four levels:

   • take action without asking
   • take action when you say so
   • ask before taking action
   • hand it off to you

   Rules are instructions the dot tries to follow. They are not a lock. So keep the important ones strict.

   Think of it like…

   Spending limits for a new employee: small things they just do, bigger ones need your sign-off.

   See an example: A sensible starting set of rules ▸

   Try this

   Always ask me before you send, publish, buy or delete anything.

   Copy

   You’re done with this step when

   Your rules say it must ask before sending, buying or deleting.

   I did this

3. Step 10 of 11

   One dot, several places to reach it

   You can talk to the same dot in ChatGPT (typing or voice), Slack or Microsoft Teams.

   • It is the same dot with the same notes, whichever you use.
   • In Slack it only answers you unless you tell it otherwise.
   • It checks before sharing anything from a private conversation.

   Think of it like…

   One colleague you can reach by desk phone, email or chat. Same person on every line.

   You’re done with this step when

   You know where you would most like to reach your dot day to day.

   I did this

4. Step 11 of 11

   Know how to really stop it

   This catches people out. Pausing a dot only stops its current main task. To stop everything:

   1. Pause the main task.
   2. In Activity, stop each task it handed off.
   3. In Scheduled, turn off anything recurring.

   Closing the chat does not stop assigned work. Stopping does not undo what it already did. Deleting your dot removes it, but messages it sent stay sent.

   You’re done with this step when

   You know where Activity and Scheduled are, and how to stop a recurring job.

   I did this

Worth knowing

• It can make mistakes even with its checks. Keep "ask before" on for anything that sends, spends or deletes.
• Your ChatGPT data settings apply to your dot. Check Data Controls if you do not want your chats used to improve models.
• Some websites block its cloud browser, so a few jobs will still need you.

Where this comes from

• Dots overview
• Getting started with your dot
• Controls and custom rules
• Computers and apps
• Tasks and memory
• Channels

Checked against the sources on 2026-10-08. These tools change quickly.