Dots are OpenAI's new always-on agents in ChatGPT, announced at DevDay on 29 September 2026. Each one runs on its own cloud computer, keeps working while you are offline, and connects to other apps through plugins. For marketers, the sensible starting point is delegating monitoring, research and first drafts, while keeping anything public-facing, financial or reputational behind human approval. This is as of 30 September 2026, and I have not run Dots myself, so this is a delegation framework built from what OpenAI and reporters have published.
Key Takeaways
- Dots are included in Pro and Business Premium plans at no extra charge, per reporting on OpenAI's announcement; Enterprise, Edu and Healthcare get a beta that admins must switch on.
- Reported details: Dots run on GPT-6 Astra, have an isolated cloud computer and browser, and connect to more than 4,000 apps via plugins.
- Proactive background research is read-only by design; sending messages or changing content needs approval, and users can set custom allow, block or approve rules.
- Best early marketing jobs: competitor and SERP monitoring, weekly reporting drafts, research briefs and inbox triage. Worst: publishing, paid spend, client-facing sends.
- OpenAI itself warns dots can make mistakes and consequential work should be reviewed.
- Pro users in the EEA, Switzerland and the UK are reported as excluded for now.
What Dots are, according to the sources
OpenAI's DevDay 2026 recap describes Dots as agents that handle ongoing responsibilities and adapt to your priorities. It lists availability as Pro and Business Premium in eligible markets, with a beta for Enterprise, Edu and Healthcare that is admin-controlled and off by default.
The more detailed description comes from The Next Web's coverage. It reports that each dot has its own isolated cloud computer and browser, can work on several projects at once, runs on GPT-6 Astra, and can be reached through ChatGPT on desktop, web and mobile, plus Slack, Teams and text messaging. Conversations with a dot reportedly do not count against your ChatGPT limits, although tasks that run in Codex or ChatGPT Work do.
I am relying on that secondary reporting for the operational details. Check OpenAI's own help pages before you build a process around them.
The safety model matters more than the features
Any always-on agent raises one question: what can it do without asking? As reported, the design has several layers:
- Background research uses read-only access to connected apps.
- Sending messages or altering content requires your approval.
- Passwords are handled without being exposed to the model, and sensitive actions such as password changes require approval.
- You can write Custom Rules to allow, block or require approval for specific actions.
- OpenAI says monitoring systems can pause or stop a dot if safety concerns arise.
That is a sensible baseline, but I would still treat approvals as the actual control, not a formality. An agent that asks permission fifty times a day trains you to click "approve" without reading.
A delegation framework for marketing work
I sort tasks along two axes: how reversible the action is, and how visible the result is. Reversible and private is the safest to delegate. Irreversible and public is the riskiest.
| Task | Reversible? | Public? | Delegate to a dot? |
|---|---|---|---|
| Weekly competitor content scan | Yes | No | Yes, good first task |
| SERP and ranking change monitoring | Yes | No | Yes, verify against your own data |
| Draft weekly performance summary | Yes | No | Yes, with human review before sharing |
| Research brief for a new topic | Yes | No | Yes, check every citation |
| Inbox triage and reply drafts | Partly | No until sent | Draft only; you send |
| Publishing a blog post | No | Yes | Not yet |
| Posting to LinkedIn or Instagram | No | Yes | Not yet; schedule yourself |
| Changing ad budgets | No | Costly | No |
| Outreach to journalists or clients | No | Yes | Draft only |
Tier 1: delegate now
Monitoring. A dot can check competitor blogs, changelogs and pricing pages on a schedule and send you a short digest. Failure mode is low: a missed item costs you little.
Research briefs. Ask for a source-cited brief on a topic before you write. The risk is fabricated or misattributed sources, so open the links. I do this with every AI-assisted brief.
Reporting drafts. Pull from connected analytics or spreadsheets, produce a first pass, and you edit. Confirm the numbers against the source, because an agent summarising a dashboard can misread a filter.
Tier 2: draft only
Emails, social copy, outreach and content updates are all fine as drafts. The rule I would set: a dot may write into a draft folder, never send.
Tier 3: keep human
Anything that spends money, publishes publicly, or speaks in your name to a client stays with you. That is not distrust of the technology; it is that mistakes here are expensive and hard to undo.
Setting up a first dot safely
- Pick one job with a clear output. For example, "every Monday, summarise new posts from these five competitors and flag topics we have not covered."
- Give it read-only access to the fewest apps needed. Do not connect your whole workspace on day one.
- Write Custom Rules before it starts. Block sending, deleting and posting. Require approval for anything account-affecting.
- Review its cloud computer's work for two weeks. OpenAI says you can open a dot's computer to see what it did; use that.
- Score accuracy. Keep a simple log: correct, partly wrong, wrong. If wrong exceeds a few percent, tighten the brief.
- Only then add a second job.
Where Dots could fail marketers
- Confident wrong summaries. Agents summarise what they see, including stale pages and misleading dashboards.
- Over-permissioned plugins. With 4,000-plus apps reachable, it is tempting to connect too much.
- Approval fatigue. Constant prompts make you less careful.
- Brand voice drift. Drafts made unattended may not match your tone; a written style guide helps.
- Regional gaps. If you are a Pro user in the EEA, Switzerland or the UK, reporting says you are excluded for now.
- Client confidentiality. If you handle client data, check your contracts before giving any agent access.
Dots versus Team Tasks
DevDay also introduced Team Tasks, which OpenAI describes as delegating recurring work using connected tools on schedules or event triggers, for Business and Enterprise plans. The practical difference: Team Tasks looks like shared team automation, while Dots looks like a personal, persistent assistant with its own computer. I cover the team-side workflow in post 355.
What this means for small teams and freelancers
For a solo marketer, a dot is closest to a junior assistant that never sleeps but needs very explicit instructions. That is genuinely useful for monitoring and first drafts. It is not a replacement for judgement about what to publish or which client to chase.
In my own work, I would use it first for the boring, high-frequency jobs: checking Search Console exports for anomalies, drafting a Monday brief, and scanning competitor pages. Those free up time for strategy and writing. The moment a task touches a client's public account, I keep my hands on it.
Sources checked
- OpenAI, DevDay 2026 Recap (availability, description).
- The Next Web, OpenAI launches dots (details on the cloud computer, safeguards, app count; secondary source).
- CNBC, DevDay 2026 live updates (context).
FAQ
What are ChatGPT Dots?
Dots are always-on agents inside ChatGPT that work on ongoing responsibilities. Reporting says each has its own isolated cloud computer and browser and runs on GPT-6 Astra. They were announced at DevDay on 29 September 2026.
Who can use Dots?
OpenAI lists Pro and Business Premium plans in eligible markets, plus an admin-controlled beta for Enterprise, Edu and Healthcare. Reporting says Pro users in the EEA, Switzerland and the UK are excluded for now. Check your account for current availability.
Do Dots cost extra?
According to the reporting I reviewed, Dots are included in Pro and Business Premium with no extra charge. Tasks that run in Codex or ChatGPT Work may count toward usage limits, so confirm the details in your plan.
Can a Dot post to social media or send emails on its own?
As described, background research is read-only and actions such as sending messages need your approval. You can also set rules to allow, block or require approval. I would keep sending and posting blocked at first.
What marketing tasks are safest to delegate?
Monitoring competitors, drafting reports, producing research briefs and triaging inboxes. These are reversible and private. Publishing, paid spend and outreach should stay under human control.
Can Dots replace a marketing assistant?
Not fully. They can take on repetitive monitoring and drafting, but they need clear instructions and human review. OpenAI itself says dots can make mistakes and consequential work should be checked.
How do Dots connect to my tools?
Through plugins. Reporting says more than 4,000 apps are reachable. Grant the minimum access each job needs, and start with read-only permissions.
Are Dots the same as Team Tasks?
No. Team Tasks is described as recurring, scheduled or event-triggered delegation for Business and Enterprise teams. Dots are persistent agents with their own computers. They can overlap in use.
Is my data safe with a Dot?
OpenAI says dots run in an isolated environment and that passwords are not exposed to the model. That does not replace checking your client contracts and data policies before connecting sensitive systems.
Work with me
I have 4+ years of marketing experience across SEO, content and organic social growth, including work with edtech brands such as Masai School. If you want a practical view of which AI agents belong in your marketing workflow, see my work and reach me through the contact form at younusfardeen.in.