To use dots, you open ChatGPT, create your dot, connect the apps it is allowed to touch, and then hand it ongoing work instead of one-off prompts. OpenAI's announcement, dated September 29, 2026, describes dots as always-on agents with their own cloud computer and browser, and says its plugins connect them to over 4,000 apps. Setup is the quick part. Deciding what the dot may do without asking you takes longer, and it is the part that decides whether the thing helps or makes a mess.
What follows is what OpenAI has actually published, checked against its announcement and Help Center on November 4, 2026, plus how we would set one up for a small marketing team.
Key Takeaways
- A dot keeps working between your conversations, on its own cloud computer, with whatever apps you connect. Pro and Business Premium plans get the first one included, in eligible markets.
- Spend your setup time on Custom Rules.
- Give it a few weeks of read-only research and drafts before it publishes anything or writes to a customer.
What OpenAI dots are
OpenAI dots are persistent agents inside ChatGPT. According to OpenAI's launch post, each one is powered by GPT-6 Astra, learns from your feedback, and keeps going while you're offline, so a Monday report can be waiting when you open your laptop. An update added to the same post on November 3, 2026 says dots can now also use GPT-6.1 Terra, picking different models for different parts of a job. Expect that list to keep changing.
The practical difference from regular ChatGPT is the computer. A dot has its own cloud machine and browser, so it can open sites, use connected apps and keep working while you're in a meeting. You can open that computer at any time and watch what it did. You can also give it permission to use your laptop, though by default your machine stays separate.
It also keeps several projects going at once. OpenAI's description is that you keep handing it tasks without juggling separate threads.
There's no new app to learn. OpenAI lists ChatGPT on desktop, web and mobile, plus Slack and Teams, and says texting is coming soon. You can message it or call it, and it can message you when it needs a decision.
How to use dots, step by step
OpenAI's Getting started with your dot page covers setup. In short:
- Create your dot in the ChatGPT desktop app, on desktop web, or in the iOS or Android app. Mobile web isn't supported.
- Name it if you like. The default handle is @yourname-dot.
- Connect plugins for the apps it should use, such as email, calendar and files. You can skip this and add them later.
- Let it introduce itself. It starts from your ChatGPT memory and the apps you connected, then suggests where it can help.
- Set Custom Rules for anything it could share, buy or change.
- Give it a standing goal, with sources, and say which decisions need you.
Step six is where most people will go wrong. A dot is built for work that continues ("keep our offsite on track, flag anything that needs me") more than for a single answer. If you only ever ask it questions, you're paying for a computer you never let it use.
Custom Rules and approvals
This is the setting we would spend the most time on. The Help Center lists four behaviors you can assign to an action: take action without asking, take action if pre-approved, ask before taking action, or hand off to you. "Pre-approved" means you asked for that exact action in your prompt. Asking for a draft does not authorize sending it.
Some lines can't be moved. OpenAI's dots privacy and safety FAQ says the most sensitive actions, like changing a password or transferring money, always require you to take over. Permanently deleting data or installing software may need approval each time. A separate auto-review checks any action that could affect your accounts or share information against your instructions and rules before it runs.
There's also a quiet mode. When you aren't working with it, a dot does what OpenAI calls proactive research: it reads your connected apps looking for ways to help, using tools that can't send messages, change content or control a browser. Anything it then wants to do goes through the normal rules.
"Dots can still make mistakes, so always review consequential work."
OpenAI, Introducing dots, September 29, 2026
Who can use dots right now
Access depends on plan and country, and the details below come from OpenAI's Help Center as of November 4, 2026. Pro users aged 18 and older can use dots in markets outside the European Economic Area, Switzerland and the UK. Business Premium users can use them in supported ChatGPT regions, subject to workspace eligibility.
Enterprise, Edu and Healthcare workspaces are in an admin preview from November 4 to November 18, 2026, during which an admin has to enable dots for your role. From November 18 they're on by default unless the admin opts out. Workspaces with UAE data or inference residency can't use them, and each member gets one personal dot.
On cost, the Help Center and launch post agree: your first dot is included in Pro or Business Premium at no extra charge, and chatting with it doesn't count toward your ChatGPT usage limits. Tasks it starts in Codex or ChatGPT Work do count, as usual.
Dots use cases for a small business
OpenAI's own examples are specific, which helps. A developer's dot reads customer feedback, builds and tests small fixes, and brings back pull requests with videos. A launch lead's dot revises launch materials when the product scope changes. A scientist's dot reruns analyses as new data arrives. A content creator's dot takes an interview transcript and prepares clips, show notes and social drafts for approval. Outside OpenAI, an early tester's dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it once he approved.
Here is how we'd map that onto a marketing team, and where we'd set the approval line.
| Job | What the dot does | Rule we'd set |
|---|---|---|
| Search and competitor watch | Reads Search Console exports, rankings and competitor pages, then summarizes what moved this week | Take action without asking (read-only) |
| Blog and page drafts | Drafts from your approved facts file, suggests titles, meta descriptions and internal links | Ask before taking action on anything published |
| Social posts | Turns one approved article into separate LinkedIn, Facebook and Instagram versions | Ask before taking action |
| Listings and citations | Compares your name, address and phone across directories against your source file and logs mismatches | Hand off to you for any account change |
| Replies to customers or leads | Drafts replies in your voice | Ask before taking action, every time |
| Ad budgets and payments | Flags pacing problems in a report | Hand off to you |
The top row is the safe place to start, because research can't embarrass you in public. Here is the first week we would run. Day one, connect Search Console exports and a shared folder, nothing else, and write the facts file described below. Ask for a Monday summary: queries that gained or lost clicks, pages that slipped, competitor pages that changed. Read it against your own dashboard. If its numbers match yours for three Mondays running, give it the drafting jobs in rows two and three. If they don't, you've lost three short reports and learned something cheap.
Give it a source of truth before you give it a goal
AI gets facts wrong without sounding unsure. Drafts state wrong prices with total confidence, so check every draft against the business's own price page. Agents make the same kind of mistake with a phone number from an old directory listing or a service the business stopped offering. The model finds a plausible answer, and nobody has told it which source wins.
So before a dot touches marketing work, write one document it reads first. Company name, address and phone exactly as they appear on Google. The services you sell, with the ones you don't. Service areas. Approved claims, prices if you publish them, and the review count as it stands today. Brand rules, including words you never use. Then add one instruction: if a fact isn't in this file, stop and ask.
That file does double duty. The same clean facts are what AI assistants look for when they decide whether to recommend you, which is the core of our AI search optimization work.
Where we would not use a dot yet
Anything that spends money without a person looking. Anything that sends a message to a customer list. Anything legal or regulated, where a wrong sentence has a cost beyond an awkward correction. The launch post also describes "specialist dots" with their own identity and access for jobs like procurement, invoice processing, email marketing and customer support. Those start as focused enterprise pilots set up with OpenAI's engineering teams, which leaves most five-person companies out for now.
We'd also be careful with volume. An agent can produce fifty articles a week. Fifty thin articles will do less for a local business than four good ones, and Google's spam policies name using generative AI tools to produce many pages without adding value for users as scaled content abuse. Our guide to AI SEO tools quotes that policy and shows where automation stops helping a site.
What dots mean for being found
Your customers are getting agents too. A dot that can browse, compare and book will research local services on its owner's behalf, and it will lean on whatever it can read clearly: your site, your Google Business Profile, your reviews and the directories that list you. If those disagree with each other, an agent has no reason to trust any of them.
So clean, consistent business data pays off whether or not you ever create a dot yourself. If you want to see how AI assistants describe your business today, our free AI visibility audit checks it, and our AI SEO team fixes what the audit finds.
Frequently Asked Questions
What are OpenAI dots?
Dots are always-on AI agents from OpenAI, launched in ChatGPT on September 29, 2026. Each dot has its own cloud computer and browser, connects to apps through plugins, and works toward goals you set, asking for approval before sensitive actions.
How do I use dots in ChatGPT?
Create your dot in the ChatGPT desktop app, on desktop web or in the mobile app, connect the apps it should use, and set Custom Rules for what it may do alone. Then give it an ongoing goal with sources, and review what it brings back.
Can you give me an example of a dot at work?
OpenAI gives several. One example is a content creator's dot that takes a new interview transcript, finds moments for clips, prepares show notes and drafts social posts for approval. Another early tester's dot noticed a missed invoice, prepared it and sent it once he approved.
How do I get access to OpenAI dots?
You need a ChatGPT Pro or Business Premium plan in an eligible market, or an Enterprise, Edu or Healthcare workspace where the admin has enabled them. Pro access excludes the European Economic Area, Switzerland and the UK, and you must be 18 or older.
Do dots cost extra?
Not on Pro or Business Premium: your first dot is included at no extra cost, and conversations with it don't count toward your usage limits. Tasks it starts in Codex or ChatGPT Work count toward those limits as usual.
Can a dot act without my approval?
Only for actions you allow. Custom Rules let you mark an action as allowed, pre-approved, approval required or handed off to you, and an auto-review checks risky actions before they run. Changing a password or transferring money always stays with you.