AI Marketing

How to Use ChatGPT Dots starts with one important limitation: you currently create a dot from ChatGPT on desktop, not from mobile.

As of October 3, 2026, OpenAI says creation is available in the ChatGPT desktop app and on desktop web. Mobile web does not support dots, and the mobile app can be used for conversation after creation when mobile access is available.

That is the mechanical requirement. The more important requirement is operational: decide what job the dot owns before you connect apps or grant permissions.

OpenAI’s official setup guide makes it easy to create and name an agent. But a business can still get a poor result from a technically correct setup if the assignment is vague.

For the related workflow, explore the persistent-agent overview.

“Help with marketing” is not a job.

“Every weekday morning, review these approved competitors and sources, summarize material changes, flag anything that affects our offers or campaigns, and return a five-point brief by 8:00 a.m. Do not post, email, or change any external system without approval” is much closer to a job.

Before you create a dot, decide what job it owns

The best starting point for a dot is a defined responsibility with clear success criteria, not an instruction to “help with everything.”

Start with the output. What should exist when the work is done? A briefing? A prioritized list? A draft? A set of updated records? A monitored queue?

Then define the source boundary. Which websites, folders, inboxes, calendars, documents, or applications are authoritative? If the agent sees conflicting information, which source wins?

Next, define the action boundary. What can it read? What can it draft? What can it change? What must it ask about every time?

Finally, define escalation. What conditions should cause the dot to stop and ask a human instead of improvising?

This turns setup into process design rather than feature exploration. If your business has never documented an AI workflow before, our AI workflow stack provides a useful framework for separating strategy, tools, execution, review and business outcomes.

How to set up your first dot

Open ChatGPT on desktop web or the desktop app and follow the Dot onboarding flow available to your account. Name the dot in a way that describes its lane. “Competitive Research” is more useful operationally than “Steve” if several people may interact with the system later.

Then give it the initial job description.

How to Use ChatGPT Dots effectively becomes much easier when the first setup gives the agent one narrow job instead of a broad company-wide mandate.

During setup, capture screenshots from the current live interface. Launch interfaces change quickly, so this article should show what a user will actually see at publication rather than relying on announcement screenshots.

OpenAI says the dot profile can show work under In progress, Scheduled, and Completed. The desktop app can also expose Activity View so the user can follow ongoing or delegated work and correct direction.

Connect only the apps the job actually needs

OpenAI says dots can use supported plugins and connections within the permissions you have already granted across ChatGPT, Work, and Codex.

That makes setup convenient, but it also creates a reason to audit access before you delegate.

A competitor-research agent may need browser research and perhaps a folder containing positioning documents. It probably does not need billing access, full email send privileges, or the ability to delete files.

A lead-triage agent may need to read a shared inbox and CRM data. Whether it should write to the CRM or send external messages is a separate decision.

Treat every connection as a capability boundary. Give the dot the minimum access needed for the current workflow, then expand only when the output has proven reliable.

For a broader access-control checklist, see our guide on how to provide access to website and marketing platforms. The same principle applies: separate accounts, correct roles, least privilege, and removal when access is no longer needed.

Write the goal like you were onboarding an employee

A reusable setup template is:

Your responsibility is: define the recurring outcome.
Use these sources: list authoritative data, documents, apps, and websites.
Never do: identify prohibited actions and high-risk areas.
Ask me before: define actions requiring human approval.
Report back when: establish cadence and escalation conditions.
Success means: define measurable quality, completeness, and timing.

The quality of that specification matters because an agent will encounter edge cases. A vague instruction forces more interpretation. A structured instruction gives the agent a standard against which it can decide whether to proceed or escalate.

Keep the first version short enough to audit. A twenty-page operating manual is not necessarily safer if nobody can verify whether the critical rules conflict.

Set Custom Rules before you increase autonomy

OpenAI’s Dots safety documentation says Custom Rules can specify whether supported actions may happen independently, require pre-approval, require approval at the moment of action, or should be handed to the user.

Custom Rules do not eliminate the platform’s built-in safety requirements, and they do not disable the separate Auto-review system.

That is useful because business autonomy should be layered.

For example, a dot could be allowed to read approved analytics without asking. It could be allowed to create a draft report automatically. Publishing that report externally could require approval. Changing account access or completing certain sensitive actions should remain human-controlled.

The goal is not to remove friction everywhere. The goal is to put friction exactly where mistakes become expensive.

Give it one real business workflow

A feature demonstration is not enough. Test the dot on work that actually matters but is still safe to review.

Workflow 1: recurring competitor intelligence. Give it an approved competitor list, the exact signals to watch, and a fixed reporting format. Require source links. Do not allow it to make public changes.

Workflow 2: content-research preparation. Have it gather primary sources, recent announcements, existing internal content, and likely cannibalization risks before a human strategist makes the editorial decision.

Workflow 3: lead or inbox triage. Let it categorize messages, identify missing information, prepare response drafts, and surface urgent items. Keep sending or CRM-changing permissions gated until accuracy is proven.

If the dot cannot perform a narrow workflow consistently, giving it more access will not solve the underlying problem.

Our AI agents for business guide shows why specialization, coordination and independent QA often matter more than simply adding more agents.

Watch its work before you stop watching closely

OpenAI says Activity View lets users inspect ongoing and delegated tasks. Use it.

The first several runs should be treated like supervised onboarding. Check whether the dot used the right source, understood priority correctly, respected boundaries, and produced an output you can act on.

When something goes wrong, correct the rule or source rather than simply fixing the final answer. Otherwise the same process defect can repeat.

A useful review asks four questions: Was the source correct? Was the reasoning consistent with the job? Did the agent stay inside its permissions? Was the output delivered in the format and timing required?

Only after those answers are consistently good should you expand autonomy.

What you can do on mobile—and what you cannot

As of October 3, 2026, the most important mobile limitation is creation: you cannot currently create a dot on mobile, and mobile web does not support dots.

OpenAI says the mobile app can support conversations with an existing dot when mobile access is available. Because rollout is gradual, publish screenshots only from a verified account and do not promise an interface the publisher has not actually confirmed.

For businesses, the practical takeaway is that initial setup and serious configuration belong on desktop.

Put the agent behind a business process

If you are building AI into your operations, your public website should be built with the same discipline: clear job, clear user path, clear measurement, and controlled complexity. Elite Web Professionals designs websites around business outcomes rather than isolated visual trends. Review our Website Design service or contact us to discuss the customer-facing system around your AI-enabled operation.

How to Use ChatGPT Dots safely comes down to starting with one clear job, limited access, explicit approval rules, and enough supervision to learn where the agent performs reliably. The fastest way to get value is not maximum autonomy on day one; it is a narrow process you can measure, correct, and then expand.

Frequently Asked Questions

How do you create a ChatGPT Dot?

As of October 3, 2026, create a dot in ChatGPT on desktop web or in the desktop app and follow the onboarding prompts available to your account.

What apps should I connect to a Dot?

Connect only the supported apps required for the job. Use least-privilege access and avoid granting unrelated accounts or write permissions simply because they are available.

What are Custom Rules for ChatGPT Dots?

Custom Rules let you add boundaries around supported actions, including whether the dot can act independently, needs approval, or should hand an action back to you.

Can a ChatGPT Dot run recurring work?

Yes. OpenAI says a dot can run reminders and recurring checks, while Team Tasks provide a separate team-owned model for shared recurring work.

Sources

• Official source: https://help.openai.com/en/articles/20001530-getting-started-with-your-dot
• OpenAI safety: https://help.openai.com/en/articles/20001529-dots-privacy-security-and-safety-faqs
• Related internal: https://www.advertisingbusiness.org/safe-way-to-use-codex-ai-computer-agents/
• Related internal: https://www.advertisingbusiness.org/how-to-provide-access/