What is ChatGPT Dots, and when should you use it instead of Chat, Work, or Codex?
Running a project takes more than writing reports or changing code. Someone has to check for new feedback, remember who has not replied, and tell people when requirements change. ChatGPT Dots is designed to take on some of that follow-up work.
You can give your Dot an ongoing responsibility: update a summary when new material arrives, investigate a stalled task, and ask you when a decision needs your judgment. It can use tools and delegate specific jobs to Work or Codex. How much effort it saves depends on the brief, the information available, and its permissions.
Information checked on October 16, 2026. The examples below come from public user reports; we have not independently reproduced them.
AI-generated concept illustration, not a product screenshot.
What is ChatGPT Dots?
Dots is OpenAI's always-on agent feature. The assistant you create is your Dot. According to the official introduction, it is powered by GPT-6 Astra and has its own cloud computer and browser for research, files, data analysis, and software work.
Suppose you are preparing a product launch. You would normally arrange competitor research, launch copy, and presentation materials separately, then revisit them when requirements change. You could give your Dot a broader brief:
For the next two weeks, help me follow this product launch. Watch for requirement changes in the project documents and the specified channel. Keep a list of unresolved issues. If a change affects the launch date or promotional materials, let me know and prepare the revisions for my review.
This is a suggested workflow. Your Dot can take in new information and adjust subsequent work, but you still need to review results. Always-on does not mean the model computes every second: it can pause and resume when a scheduled time or the conditions for further work arrive.
Choosing between Chat, Work, Codex, and Dots
Start with what you need to get done.
| Experience | Best suited to | Example |
|---|---|---|
| Chat | Questions, discussion, explanations, and drafting | “Explain this email and help me decide how to reply.” |
| ChatGPT Work | Research, analysis, reports, spreadsheets, and scheduled tasks | “Compare these quotes in a table and explain your recommendation.” |
| Codex | Writing, debugging, testing, and reviewing code | “Fix the login page, run the tests, and show me the changes.” |
| Dots | Ongoing follow-up, coordinating tasks, and responding to changes | “Track feedback on this project, arrange investigations, and ask when I need to decide.” |
This follows OpenAI's comparison. Work already supports multistep assignments and scheduled work, as its getting-started guide explains. The reason to choose Dots is that you also want someone to keep following up.
For a single article edit, a Chat conversation may be enough. For a research report, use Work. If you already know which software bug needs fixing, go straight to Codex. A Dot is useful for the questions that remain afterward: which feedback is still unresolved, whether the last fix worked, and what should happen next.
On a software project, your Dot can collect feedback and organize investigation while Codex handles the code and you review the changes. For a closer look at coding tools, see our Claude Code vs. Codex comparison.
How does it keep following up?
The public documentation describes a combination of memory, execution environments, and task scheduling. This explains the product's behavior; OpenAI has not disclosed its complete internal implementation.
Carrying context into the next task
Your Dot uses the current conversation, relevant ChatGPT memory, and its own saved work notes. These notes can preserve preferences, decisions, and unfinished work.
Memory is not a complete chat archive that every task can see. A delegated task has its own conversation and receives the instructions and context it needs. Give the task important files, versions, and acceptance criteria explicitly. See the tasks and memory documentation, or our guide to clearing ChatGPT memory.
Working in the cloud and accessing apps through plugins
Your Dot's cloud computer has its own files, software, and browser sessions. Cloud tasks can continue after your computer shuts down. To use email, calendars, or repositories, connect the relevant plugins and grant the necessary permissions.
The cloud browser does not inherit your local browser's signed-in sessions. When a website needs a login, use the private sign-in flow or take over the browser. You can also connect your own computer for local files and apps; it must stay online with the ChatGPT app open. See computers and app connections.
Scheduling work and resuming when needed
A Dot can split work among background agents or create Work and Codex tasks. You can keep talking to it, add instructions, or change priorities while they run.
Checks at fixed times need a saved schedule. For “check the project channel every day at 9 a.m. Beijing time,” specify what to look for and when to notify you. Adding a Dot to a channel does not automatically establish monitoring. Ask it to confirm the saved schedule, then check the time and scope. See recurring tasks.
Your Dot can also research information it has permission to read. That proactive research is read-only; editing files or sending messages afterward still requires the applicable permissions. Actions are reviewed against your instructions and permissions, with approval or a handoff when necessary. Asking for an email draft does not authorize sending it. See controls and permissions.
Which plans can use Dots?
As of October 16, 2026, the official eligibility list says:
| Plan | Availability |
|---|---|
| Pro 100, Pro 200, Pro 500 | Gradual rollout to users over 18 outside the European Economic Area, UK, and Switzerland |
| Business Premium | Rolling out worldwide |
| Enterprise | Rolling out worldwide; off by default and requires an administrator to enable it |
| Free, Go, Plus, standard Business, Edu | Not separately included in this eligibility list; access cannot be confirmed from it |
Distinguish standard Business from Business Premium before buying. Access to Work or Codex does not establish Dots eligibility, and eligible accounts may still be waiting for rollout.
You can create your Dot in the ChatGPT mobile app, desktop app, or desktop browser. Mobile web is not supported. Update the mobile app and follow the getting-started guide.
How do Dots usage limits work?
Three rules are confirmed. Together, they do not amount to unlimited use.
| Activity | Usage treatment |
|---|---|
| Talking directly to your Dot | Does not count toward ChatGPT usage limits |
| Your Dot doing deeper work | Your plan includes an allowance, with extended limits for the first month after launch |
| Work or Codex tasks started or managed by your Dot | Count against those products' limits as usual |
These rules come from the official Dots documentation. It does not give a universal formula for points per message or hours per month. The first-month extension is not a permanent commitment.
A short request can trigger a lot of work. “Organize these PDFs” may involve reading files, searching, reasoning, and several revisions. Work and Codex share usage, which also depends on the model, context, tools, and caching. A short prompt is not necessarily cheap, and API dollar prices cannot directly tell you how many tasks your subscription includes. See the usage documentation.
Use the official usage-dashboard instructions to check remaining allowance and reset times. Run one bounded task first and compare usage before and after. Some users report background tasks consuming their weekly allowance quickly, but their percentages cannot be applied to every account.
What have people actually used Dots for?
Public reports include both useful workflows and jobs that needed repeated correction. These are users' own accounts, useful for ideas rather than guarantees of reliability.
Grocery offers followed by scheduling a shopping trip
Reddit user Northern_Virginia describes a household workflow: every Thursday, their Dot checks local grocery offers against an approved shopping list and emails the results to their wife. Once she decides what she wants, it checks the user's calendar for a convenient shopping time the following week. See the original post.
There is a defined list, a regular check, and a decision to wait for before proceeding. The follow-up matters because finding prices is only one step. Email and calendar connections and authorization for the relevant actions are still needed.
Reviewing a large PDF collection while taking side questions
In an October 1 Reddit post, a user described reviewing hundreds of PDFs containing thousands of pages. One review continued for days, with questions inserted along the way before returning to the original task. See the review report.
Some explanations of the execution environment came from the Dot itself, and commenters questioned whether they had been verified. Treat this as a workflow report, not proof of coverage or accuracy. Ask for a file inventory, page references, and a small initial batch so you can spot omissions.
Product updates, websites, and content work
AI Agents Library's Diana tested ten business tasks with a Dot named Sparky. Website creation and publication, product research, and content planning went relatively well. One scheduled information check also ran successfully and sent a phone notification.
Deliverable details were less reliable: spreadsheet formulas broke after importing into Google Sheets, and inbox triage needed supervision. The test covered two days and only one scheduled run, so it did not establish reliability over several weeks. See the full review.
Content creators could start by tracking a few official sources and collecting update cards. Once the information quality and notification frequency work well, website maintenance or draft preparation can follow.
Handling email and assigning small jobs during calls
In an October 7 post, note author yuta said their Dot used earlier discussions to help process email, compare options, and even identify a reservation error. They also liked assigning something during a voice conversation and continuing to discuss other matters.
Complex work was harder: casual requests often needed more detail, and the author opened delegated tasks to check their direction. See the original account.
Other early users reported slow actions and browser or computer-use obstacles, while some already found reminders, notes, and email follow-up helpful. See the mixed Reddit feedback. The practical question is whether the time saved on checking and organizing exceeds the time spent correcting and explaining.
What should you delegate first?
Choose something you already check every week. For blog research, you could adapt this brief:
For the next four weeks, follow these official sources every weekday at 9 a.m. Beijing time. When you find a product launch, pricing change, or important feature update, prepare a short card with the change, original link, possible article angle, and questions still needing verification. Notify me only when there is a worthwhile topic. You can research and prepare drafts; ask me before publishing. Confirm the saved schedule, then run one check.
Check whether the first result uses the right sources, mistakes old news for new developments, or misses eligibility conditions. Tighten notification criteria if there are too many updates, and specify what is missing if the research is thin.
For software work, your Dot can follow feedback and pass defined issues to Codex. The October 5–9 update added details about using existing Codex threads and automation context, so account for version differences when reading early reviews.
Use Activity to inspect progress and files. Stopping your Dot's main task, delegated tasks, and recurring schedules requires separate actions; ending a call does not cancel the work. See the work controls.
After a week or two, check whether it remembers unfinished work, contacts you at useful moments, and produces results that need little revision. Those observations will tell you more than the always-on label about whether it belongs in your workflow.
Sources
- OpenAI — Meet dots
- OpenAI — Get started with ChatGPT Work
- OpenAI — Get started with your dot
- OpenAI — Tasks and memory
- OpenAI — Connect computers and apps to your dot
- OpenAI — Control your dot
- OpenAI — Pricing
- OpenAI — Create your dot in the ChatGPT mobile app (October 5–9, 2026)
- yuta — I Tried Using ChatGPT Dots (2026-10-07)
- Reddit — I spent a day poking Dots with sticks (2026-10-01)
- Reddit — ChatGPT Dots so far (2026-09-30)
- Reddit — What are the coolest use cases for your dots?
- AI Agents Library — I Tested OpenAI Dots on 10 Real Business Tasks