top of page

What Are ChatGPT Dots? How They Work and How to Get Started

Oct 1
9 min read

Updated: 2 days ago


By Ben Angel · Sources checked November 11, 2026


ChatGPT dots are always-on AI agents that can keep working on a goal between conversations. Each dot has its own cloud computer, can use apps you connect, and brings back results or questions when it needs your input. OpenAI introduced dots on September 29, 2026. OpenAI’s announcement


For a solopreneur, recording a video is only the start. Pulling out useful moments, drafting posts, and checking everything can swallow the rest of your afternoon.


Here’s what dots can help with, how to get started, and how to keep control of the work.


In this guide



What are ChatGPT dots?


OpenAI dots introduction page with a yellow character, a GPT-6 Astra description, and buttons to create a dot or download ChatGPT.
OpenAI’s dots introduction page. The capture shows official promotional material; access and setup details can change during rollout. Source: OpenAI.

A dot is an AI assistant you can give an ongoing responsibility, with boundaries you set.


Dots can draw on different models for different parts of the work. They can help with research, documents, data analysis, and software work, using available context and tools. You can continue the conversation while work is underway. Meet dots


How do ChatGPT dots work?


There are three practical pieces to understand:


  • The goal: Tell your dot what you want finished, which sources matter, and what a good result looks like.

  • The tools: Its cloud computer and browser handle supported work. Connected apps provide additional information and actions within the access you’ve granted.

  • The feedback: Review the result, point out what’s wrong, and tell it what to improve next.


Your dot’s cloud computer is separate from your laptop. Cloud work can continue while your device is off. Connecting your own computer is optional; work that needs that computer requires it to be online with the ChatGPT app open. Its cloud browser also has separate sign-in sessions. Computers and apps


You can also talk through a task by voice. OpenAI’s DevDay dots demonstration shows a short voice interaction at 11:36–11:45. It’s an official launch demo, rather than an independent test of a complete business workflow.



OpenAI’s DevDay 2026 demonstration shows a short voice interaction with a dot at 11:36–11:45. Credit: OpenAI. This launch demo does not establish the completion of an end-to-end business task.



Dots, ChatGPT Work or Codex: which should you use?

Use ChatGPT Work for a defined deliverable, such as a research brief or spreadsheet, including work on a schedule. Use Codex when the job centers on building, debugging or reviewing software. Give a dot an ongoing responsibility that needs follow-up as circumstances change. It can delegate parts of that work to Work or Codex. These experiences overlap; Work is not limited to one response or one-off jobs. See OpenAI’s comparison and ChatGPT Work guide.


What I found in my first 24 hours


I’ve been using dots for 24 hours. Here’s what I found: it’s more useful when I give it access to the skills I’ve already set up and ask it to work within existing projects. Those projects contain Markdown files, or plain-text notes, with instructions and background, so there’s less to explain from scratch.


My biggest win as a solopreneur was discussing five projects while it kept working in the background. In my use, it carried relevant context as I moved between them. The jobs included:


  • Drafting a new blog post

  • Developing video scripts

  • Creating thumbnails

  • Uploading a YouTube video

  • Preparing a media-partner video submission and organizing its links

  • Organizing files for a Google Drive upload

  • Planning a content-review dashboard


Some of that work was still in draft or preparation. The YouTube upload was completed. I personally sent the media email and uploaded the Drive file; my dot helped prepare and coordinate those steps.


That’s my early experience. Project files and skills must be available with your permission. You may need to correct or restate context, especially when work moves to a separate task.


Three ways solopreneurs can use dots


Try these small workflows with material you can verify.


1. Turn one recording into content drafts


OpenAI illustrates a creator workflow in which a dot finds potential clip moments and prepares show notes and social posts for approval. Official creator example


Try giving it one transcript, two examples of your writing, and a clear audience. Ask for three post drafts, suggested clip timestamps where the transcript includes timing, and a list of claims to check.


Your review should answer: Does this sound like me? Is the advice accurate? Has it preserved what I actually said?


Keep publishing as a separate decision. This is especially important when the recording mentions a client, a price, or an offer that may have changed.


OpenAI’s Iggy dot example showing transcript-based clip suggestions, show notes and social drafts, followed by a request to revise the content kit.
OpenAI’s Iggy example illustrates turning a transcript into content drafts and carrying feedback across a release kit. This is an official product illustration, not an independent workflow test. Source: OpenAI.

2. Prepare for a client conversation


Give your dot the specific proposal, approved notes, and questions for an upcoming call. Ask for a one-page brief covering the client’s priorities, unresolved questions, and information you still need.


For example: “Use these three documents to prepare my meeting brief. Link each important claim to its source. Flag conflicting dates or prices. Don’t contact the client or change the proposal.”


Check the brief against the originals. A polished summary is only useful if it represents the client accurately. You still own the pricing, promises, and relationship.


3. Keep a small launch organized


Try a bounded assignment: “For the next seven days, help me maintain this launch checklist. Use the documents I provide. Flag overdue decisions and draft suggested updates for my review.”


Specify who owns each task and what counts as finished. Ask the dot to separate confirmed facts from assumptions. If you want a scheduled check-in, give a time, time zone, and end date, then ask it to confirm the setup. Working with tasks


How to get started with ChatGPT dots


OpenAI’s How to get your dot panel showing eligible ChatGPT plans, rollout information, and the Create your dot button.
OpenAI’s access panel explains eligible plans and administrator requirements. Availability can change during rollout. Source: OpenAI.

Check your access and plan


Access checked November 11, 2026: eligibility depends on your plan, region and workspace. Pro and Business Premium access is rolling out. Enterprise, Edu and Healthcare have separate administrator controls and eligibility exclusions; November 18’s planned default access is still ahead. Check the current requirements before subscribing. Access details · Age eligibility


“Always on” does not mean unlimited work. Tasks delegated to Work or Codex use those products’ allowances. Check the current plans and your account’s usage limits before starting substantial work.


Set up one useful responsibility


  1. Create your dot. Use the latest ChatGPT app on iOS or Android, the desktop app, or ChatGPT in a desktop browser. Follow the setup prompts. Start with whichever supported device you already use.

  2. Connect only what the first job needs. You can skip app connections initially and add them later.

  3. Choose whether to connect your computer. Leave this for a task that needs local files or tools.

  4. Explain your business briefly. Share your audience, the task, relevant examples, and the decisions you want to make yourself.

  5. Ask for one reviewable output. A brief or a few drafts makes a manageable first test.


You can set up and talk to your dot in the ChatGPT mobile app. Mobile web is not supported. Open your dot’s profile to see the contact methods available to you. Texting setup also works in the mobile app where eligible. Check the current setup guide if your controls look different.


Permissions and review before you hand over work


OpenAI’s Action review and approvals explanation describing auto-review, Custom Rules, safety requirements, and sensitive tasks that remain with the user.
OpenAI explains how action review checks instructions, Custom Rules, and safety requirements, including actions that need your approval. Source: OpenAI.

Start by checking existing app connections. Plugin permissions are shared across ChatGPT, Work, Codex, and dots, so your dot may be able to use connections you already approved. Disconnecting a service stops new access through that connection; it doesn’t erase information already retained. Privacy and permissions FAQ


OpenAI describes several layers of protection. Auto-review checks relevant actions against your instructions and safety requirements. Custom Rules let you set ongoing boundaries, but cannot remove core safeguards. Highly sensitive steps, such as changing passwords, require you to take over. OpenAI’s safety explanation


Dots can also look for helpful updates in connected information. The tools used for that proactive research are read-only. Any resulting action still has to follow the applicable permissions and approval rules.


For a first business task, make your boundary explicit: “Prepare drafts for me. Ask before sending, publishing, spending money, or changing a live page.” Review any rule confirmation the app presents.


Also decide what information belongs in the task. Use a public article or a cleaned-up sample before handing over confidential client material. Never put passwords in the chat; use the supported secure sign-in flow when needed.


Check important work before relying on it. Look for unsupported claims, the wrong recipient, outdated numbers, and changes outside your request. You can inspect progress in Activity and redirect the work. Control your dot


Your first task to try today


OpenAI’s Give it a responsibility to own example showing in-progress and scheduled research tasks beside an orange dot character.
OpenAI’s product illustration shows a dot working through an assigned responsibility and bringing findings back for review. Start with one small, reviewable task. Source: OpenAI.

Choose one article you own and already trust. Paste this prompt, replacing the bracketed details:


“Help me repurpose this article for [audience] on [platform]. Use only the article and these two examples of my writing. Create three short post drafts with different angles. For each one, show the supporting passage from the article and flag any claim that needs checking. Keep the tone [describe it]. Don’t invent a personal story, statistic, or result. Return the drafts here for review. Don’t publish, schedule, contact anyone, or edit the original. Finish after delivering the three drafts.”


Then check the work against the source. Pick one draft and give specific feedback: “This opening is too broad. Lead with the reader’s problem, shorten the introduction, and remove the claim about saving five hours.”


Check how much editing it needs, then repeat with another article before expanding the responsibility.


For a more detailed task brief, see my guide to using Goals for business tasks. Use it to define the deliverable, checks and approval boundaries before handing over a larger assignment.


Give it a responsibility, not just an activity

Once that first test works, try a bounded ongoing assignment. For example:

For the next two weeks, help me keep my weekly video release on track. Use this transcript, my approved offer sheet and the checklist in this project. Keep a list of missing assets and decisions. Prepare copy changes as drafts and link factual claims to their sources. Tell me in this chat if a deadline is at risk; otherwise collect questions for my review. Do not send, publish, spend or change live pages. If I request a fixed check-in, confirm its time zone, end date and saved schedule before treating it as active.

Review the first result before giving it more responsibility. Check the actual output and its sources; a completed task status does not prove that the result is correct. How tasks and memory work.


OpenAI documentation example of Alfred coordinating a mobile checkout fix, with task links, Activity, Outputs and the Assigned work explanation.
OpenAI’s documentation example shows Alfred coordinating tasks, with Activity and Outputs beside the conversation. This is an official documentation illustration, not a capture of Ben’s account. Source: OpenAI, checked November 11, 2026.

Build an AI team around your business


Ben Angel smiling beside a laptop and a phone displaying ChatGPT, against a green leafy background.
Ben Angel, creator of Zero-Employee Entrepreneur, pictured with a laptop against a green background.

Once you’ve tried a small task, the next question is how the work fits together. Each AI specialist needs a clear job, useful context, a shared goal, and a process for handing work back to you.


That’s what my Zero-Employee Entrepreneur course is designed to help you build: a personalized AI team, your SoloStack, around one meaningful 90-day business goal. You work through the roles, instructions, approval boundaries, handoffs, and quality checks. Explore the course if you’re ready to turn individual AI tasks into a more coordinated way of working.


ChatGPT dots FAQ


Are ChatGPT dots free?


Access depends on an eligible plan or workspace, and delegated work has usage limits. Check OpenAI’s current access and billing guidance.


Why can't I see ChatGPT dots?


Check your plan, region and workspace eligibility first. Dots are still rolling out, and some workspaces need an administrator to enable access. On mobile, update the ChatGPT app; a phone browser is not supported. Check OpenAI's current access requirements before changing your subscription just to get dots.


Do I need to know how to code?


No coding is needed to create a dot or describe a task. Start with a straightforward written request and clear examples, following the setup guide.


Can a dot send emails or publish content?


Potentially, when the necessary tools, access, and authorization are in place. Approval requirements depend on the action. Asking for a draft does not give permission to send it. Be explicit about your boundaries and review consequential work. Action approvals


Does closing the app stop the work?


Assigned cloud work can continue. To stop work, check the controls for the main task, delegated tasks, and scheduled tasks separately. Ending a voice call doesn’t necessarily end an assignment. Stopping and reviewing work


If you’re new to the basics, start with my beginner’s guide to ChatGPT. Then bring your dot one useful job, a reliable source, and a clear definition of finished.


About the author: Ben Angel is an Entrepreneur Network contributor and author of The Wolf Is at the Door. He writes about practical AI workflows for entrepreneurs.


About this guide: This AI-assisted guide combines official documentation checked on November 11, 2026, with Ben’s account of his first 24 hours using dots.

Comments


bottom of page