How do I start using OpenAI Dots?

Open the ChatGPT desktop app or desktop browser, create your first Dot, and give it a name. Hand it an ongoing goal, describe what done looks like, and connect the few apps it needs. Your first Dot is included at no extra cost on eligible Pro and Business Premium plans in eligible markets.

What should I give a Dot first?

Give it an ongoing goal rather than a one-off question. A Dot is built to take on a project and keep working, so a task with a clear definition of done — and a way to check the result — plays to its strengths. See our Dots use cases for concrete examples.

How do I control what a Dot can do?

Use Custom Rules to allow an action, require your approval before it happens, or block it. You also choose which apps the Dot can access through ChatGPT app controls. Background proactive research uses read-only tools, and some sensitive tasks, such as changing a password, always stay with you.

How do I know what my Dot did?

The activity view shows progress, including work it did in the background. An auto-review step checks actions against your instructions, Custom Rules, and safety requirements, and you can open the Dot's own cloud computer at any time to inspect its work.

Why can't I see Dots, or why did a task stop?

If you do not see Dots at all, it is usually plan or market eligibility — check the official eligible markets page . If a task hits a usage limit, it was likely routed through Codex or ChatGPT Work, because those count toward usage limits while conversations with a Dot do not. See our Dots FAQ for more.

How to Use OpenAI Dots: A Beginner's Guide

Using a Dot is less like prompting a chatbot and more like briefing a new teammate. You create the Dot, give it a name, hand it a goal, and then work with it as it makes progress in the background. This guide walks through the whole loop: create, start, instruct, connect, run, review, and improve. If you need the concept first, read what OpenAI Dots are; if you are not set up yet, start with how to get a Dot.

What is a Dot?

A Dot is an always-on AI agent from OpenAI. It is not a single conversation — it is a named helper with its own cloud computer and browser that takes on a project and keeps working while you do other things. Dots are powered by GPT-6 Astra and can connect to 4,000+ apps through OpenAI's plugin ecosystem. You reach your Dot through ChatGPT on desktop, web, and mobile; it is also available in Slack and Teams, and it can message you first with updates. Each Dot learns from your feedback over time. For the full picture, see our explainer on what OpenAI Dots are.

Create a Dot

You create your Dot from the ChatGPT desktop app or desktop browser, where the setup flow lives. Give it a name you will recognize at a glance, because you may add more Dots later. On eligible Pro and Business Premium plans, your first Dot is included at no extra cost. Enterprise-family workspaces, including Edu and Healthcare, can use the beta once a workspace admin enables it.

The name matters more than it sounds. A Dot holds context and standing instructions, so a clear name — "Launch Assistant", "Support Triage" — makes it obvious which Dot you are talking to when several are running.

Start a Dot

The most common first mistake is treating a Dot like a search box. Instead of one prompt, hand it an ongoing goal: "keep the changelog current", "turn incoming feedback into triaged issues", "watch this dataset for changes worth flagging". A goal gives the Dot something to keep working toward, which is what its always-on design is for. If you want inspiration, our Dots use cases page lists the kinds of assignments that fit well.

Give a Dot instructions

Good instructions describe what done looks like, not just what to do. Say what the output should contain, what quality bar it must clear, and what the Dot should do when it is unsure. Then use Custom Rules to set the boundaries. Custom Rules let you:

Allow an action so the Dot can proceed on its own.
Require approval so it checks with you before acting.
Block an action outright.

This is where you encode your preferences once and let the Dot apply them consistently. Auto-review then checks the Dot's actions against your instructions, your Custom Rules, and safety requirements before they go ahead.

Connect tools

A Dot can only act in the tools you connect. You choose which apps it can access through ChatGPT app controls, drawing on the 4,000+ apps available through OpenAI's plugin ecosystem. The practical advice is to start read-only where possible.

Read first. Connect sources the Dot can read from — documents, a knowledge base, a tracker — so it can build context without changing anything.
Add one workflow. Give it a single, low-stakes goal and watch how it reports progress in the activity view.
Grant write access later. Only let a Dot send, post, or change things once you have seen it behave well on read-only work.

Run a task

Once a Dot has a goal, instructions, and tools, let it run. A Dot works in the background, including while you are away, and it can take on a project while handling several others. Crucially, it does not wait silently: a Dot can message you with progress, questions, or decisions that need your input. Treat those messages as the natural checkpoints — answer the questions and the Dot keeps going.

Review results

Always close the loop by reviewing what the Dot produced. Several controls exist for exactly this:

Activity view. See what the Dot has been doing, including background work.
Auto-review. Actions are checked against your instructions, Custom Rules, and safety requirements.
Open its computer. You can open the Dot's own cloud computer at any time to inspect its actual work.
Sensitive tasks stay with you. Some actions, such as changing a password, are always reserved for the human.

Dots can make mistakes. Review consequential or irreversible output before you rely on it, especially anything that moves money, data, or public messages.

Improve a Dot

A Dot learns from your feedback over time, so the correct response to a mediocre result is not to start over but to redirect it. Say what was wrong, what you want instead, and update the Custom Rules if the fix should be permanent. Over a few iterations the Dot gets closer to doing the work the way you would do it.

Example Dot workflows

OpenAI illustrated Dots with concrete roles. Here are three workflows drawn from those examples that show the loop end to end.

Workflow	What the Dot does
Feedback to pull requests	Watches customer feedback for recurring requests, scopes smaller fixes, builds and tests them, and brings complete PRs with attached videos showing the changes.
Interview to content	Learns your voice and audience, then turns a new interview transcript into clip moments, show notes, and draft social posts for your approval, carrying your edits across the materials.
Analysis kept current	Learns your research question and how you weigh evidence, reruns analyses as new evidence arrives, updates figures, and flags what needs your review when findings change.

Common problems

Symptom	Likely cause and fix
You do not see Dots	Almost always plan or market eligibility. Confirm your plan is Pro or Business Premium (or that an admin enabled the enterprise beta) and that your market is on OpenAI's list, then consult the how to get a Dot guide.
A task hits a usage limit	Tasks you start in Codex or ChatGPT Work count toward your usage limits, while conversations with a Dot do not. If a limit appeared, the work likely ran through one of those routes.
The Dot made a mistake	Review the output, correct the Dot, and tighten your Custom Rules so the boundary is explicit next time. Treat consequential work as something to approve before it takes effect.

Next steps

The safest path is to start small: one Dot, one goal, a couple of read-only apps, and a habit of reviewing the activity view. For quick answers to the questions people ask most, see the Dots FAQ, and for the concept behind the product, go back to what OpenAI Dots are.

Source: OpenAI's Introducing Dots announcement and Dots help center, checked 2026-09-30. This site is independent and not affiliated with OpenAI. Dots can make mistakes; review consequential work.