OpenAI Dots are designed for work that needs attention beyond a single ChatGPT conversation. Instead of asking ChatGPT to summarize your inbox once, for example, you can configure a Dot around an ongoing goal, connect the relevant tools, and define how it should handle the work.
The important part is setting it up correctly. A Dot needs a clear objective, access to the right information, and boundaries around what it can do without asking you first.
This tutorial walks through the process of creating your first Dot, connecting an app, assigning a practical task, configuring approval rules, and checking the results. It also explains what to do if Dots is missing from your account or a connected app does not work as expected.
Table of Contents
What You Need Before Setting Up a Dot
Before opening the setup screen, check three things: account access, device, and the task you want to delegate.
Check your account. OpenAI's current rollout includes Pro users in supported markets and Business Premium users in supported regions. Enterprise, Edu, and Healthcare workspaces may have access through a beta when an administrator enables it. Availability can change as the rollout expands.
Use a supported ChatGPT app or website. You can create a Dot in the latest ChatGPT mobile app for iOS or Android, the desktop app, or desktop web. Mobile web is not supported. Some connected messaging-channel setup, such as Slack or texting, may require desktop.
Choose a task with a clear outcome. Do not start by giving a new Dot unrestricted access to your inbox, files, and calendar. Pick one contained task, such as preparing a daily briefing from approved sources or summarizing updates from a specific project channel.
OpenAI's Getting started with your dot guide is the primary reference for current availability, setup, connected apps, and controls. If you want the broader product picture before following the steps, see ToolJunction's OpenAI Dots features, use cases, and pricing.
How to Create Your First OpenAI Dot
Step 1: Open ChatGPT
Sign in to the latest ChatGPT mobile app for iOS or Android, the desktop app, or ChatGPT on desktop web using the account you want to use for Dots. Mobile web is not supported.
Look for the Dot entry in the sidebar or the available Dot creation option in your account. If it is not there, do not assume you have missed a step. Access depends on your plan, region, workspace settings, and the current rollout.
If you use a managed work account, check with your administrator if Dots is not available. The workspace may need the relevant beta enabled.
Step 2: Start the onboarding process
Select the option to create a Dot and follow the onboarding prompts.
You will be asked to give your Dot a name. Choose something that makes its purpose obvious, especially if you expect to create more than one. For example, “Project Briefing” is easier to recognize than a vague name such as “Assistant.”
Complete the remaining setup prompts and open the Dot's profile once it has been created. This is where you can review its activity and available settings.
At this stage, you have created the agent, but you have not necessarily given it everything it needs to complete a task. App connections, permissions, and instructions are separate decisions.
Step 3: Decide whether it needs a connected app
Before connecting anything, write down where the Dot needs to get its information.
For example:
- A project briefing might require access to a specific Slack channel.
- A calendar summary needs the relevant calendar connection.
- An email triage task needs access to the appropriate email account.
- A research task may need only approved sources and read-only access.
This prevents a common setup mistake: connecting several apps before knowing which ones the task actually requires.
Step 4: Connect the required app
Open the Dot's profile, select Customize, then open Plugins. Choose the supported app you need. The available options can vary by account and rollout.
Select the app and follow its authorization process. Read the access request before approving it. Check which account is being connected and whether the integration supports reading information, making changes, or both.
After authorization, confirm that the app appears as connected. If the Dot needs access to a particular channel, folder, or resource, check that the account and permissions cover that resource too.
A successful connection does not mean every action is automatically available. App permissions and the integration's supported capabilities determine what the Dot can actually do.
Keep the first setup narrow. If your goal is to prepare a briefing from one project channel, connect that source first. Add another app only when the task genuinely needs it.
How to Assign Your First Task to a Dot
Once the Dot has the required access, give it a task with a defined outcome. This is where the difference between a useful agent and a vague instruction becomes clear.
Consider an operations manager who wants a morning summary of project updates. “Keep me updated” leaves too much open to interpretation. The Dot does not know which updates matter, which sources to use, or what the final briefing should contain.
A better instruction defines the goal, scope, output, and restrictions.
Copyable first-task prompt
Use the following as a starting point and replace the bracketed details with your own information.
Goal: Prepare a project briefing from approved sources.
Sources: Use only [the specific project channel or connected source] and the documents I have authorized.
Time period: Cover updates from [the relevant period].
Output: Prepare a report with four sections: important updates, supporting evidence, unresolved questions, and suggested next steps.
Boundaries:
- Do not send messages or contact anyone.
- Do not modify project records or publish the report.
- Do not invent missing owners, deadlines, or decisions.
- Flag conflicting information and anything that needs my review.
- Ask me before taking any action beyond preparing the report.
Completion criteria: The report should be concise, identify the source of important claims where possible, and clearly separate confirmed information from suggestions.
This prompt gives the Dot something specific to work toward. It also makes the result easier to evaluate because you know what should appear in the final report.
Remember that written instructions are not a substitute for technical access controls. Configure the Dot's available custom rules and connected-app permissions separately.
Run the task and inspect the first result
After submitting the task, check whether the Dot needs clarification, app authorization, or approval before it can continue.
When the result is ready, compare it with the original source. Check whether important updates are missing, whether any statements go beyond the evidence, and whether suggested next steps are presented as recommendations rather than completed actions.
If the report is too broad, refine the instructions. For example, ask the Dot to prioritize deadlines and blocked tasks, exclude routine updates, or limit the briefing to a specified length.
Do not move straight from a successful first report to unrestricted automation. First establish that the Dot can produce a useful result consistently.
How to Set Approval Rules for Your Dot
A Dot's task instructions describe what you want it to accomplish. Custom rules help define how it should handle actions along the way.
Open the Dot's profile, select Customize, and review Custom rules. OpenAI documents options that allow a Dot to act without asking, act when an action has been pre-approved, ask before taking an action, or hand the action over to the user.
The right choice depends on the consequences of getting something wrong.
| Task or action | Sensible starting approach |
|---|---|
| Summarize approved project updates | Allow the research and reporting work, then review the output |
| Draft an email | Let the Dot prepare a draft for your review |
| Send an email to a customer | Require approval before sending |
| Change a business record | Require approval and verify the target record |
| Publish content or share a file externally | Keep human approval in the process |
| Make a consequential financial or operational decision | Retain human control over the decision |
These are practical recommendations, not a guarantee that the Dot will follow every instruction perfectly. The available controls and app permissions still matter.
For a deeper explanation of least-privilege access, read ToolJunction's guide to AI agent permissions. It explains why agents should receive only the access they need and why human approval matters when an action can affect other people or business data.
How to Review a Dot's Work
Do not judge an agent only by whether it says the task is complete. Review the actual result.
Open the Dot's profile and look at its task activity. The interface can show tasks that are in progress, scheduled, or completed, along with task details and access to the Dot's computer.
Use this review process:
- Check the status. Determine whether the task is running, waiting for something, or completed.
- Open the result. Read the output rather than relying only on the status label.
- Verify the evidence. Compare important claims with the connected source or original information.
- Check for unauthorized actions. Confirm that the Dot did not send, publish, or modify anything outside the task's intended scope.
- Give specific feedback. If the result is incomplete, explain what is missing and what a correct result should include.
- Pause if necessary. If the Dot is behaving unexpectedly or its permissions need to be reviewed, use the available pause controls in its profile.
A completed task can still contain errors. This is particularly important when the output will be shared with a client, used for a decision, or treated as an official business record.
Can a Dot Work While You Are Offline?
Dots are designed to continue working toward assigned goals between conversations. They use their own cloud computer, so you do not generally need to keep your personal computer running for cloud-based work.
Local computer access is a separate, optional capability. If you configure a Dot to use your local computer, that computer must remain online with ChatGPT open for the local work to continue.
Do not enable local access just because the option is available. Use it only when the task genuinely requires resources on your computer and you understand the access involved.
OpenAI's Introducing dots announcement explains the product's always-on design, connected apps, cloud computer, and intended use cases.
Common OpenAI Dots Setup Problems
Dots does not appear in my ChatGPT account
First, confirm that you are signed in to the correct account. Check plan eligibility, regional availability, and whether the rollout has reached your account. If you use a managed workspace, ask the administrator whether Dots is enabled.
If you are using a mobile browser, switch to the latest ChatGPT app for iOS or Android, or use ChatGPT on desktop web. Mobile web is not supported for Dot creation. If the Dot option is missing, check plan eligibility, region, workspace settings, and rollout status.
My Dot cannot access a connected app
Open the connection settings and confirm that the correct account is authorized. Check whether the required resource is available to that account and whether the connection supports the action you are requesting.
If access has expired or been revoked, you may need to reconnect the app. Do not work around a permission problem by granting broader access without first understanding why the original connection failed.
The Dot keeps asking questions instead of finishing the task
Some clarification is useful when the task lacks important information or requires approval. If the Dot repeatedly asks about the same issue, make the instructions more specific.
Define the time period, source, output format, and actions it must avoid. If an approval is required, review the request rather than instructing the Dot to bypass it.
The task is complete, but the output is wrong
Give the Dot a precise correction. Identify the incorrect claim, point to the relevant source, and explain what needs to change.
For example, instead of saying “make the report better,” say: “The report includes an unconfirmed deadline. Remove it and mark the deadline as not specified unless the connected source confirms it.”
Review the revised output before using it. Repeating a task does not guarantee that the next result will be accurate.
I need to stop a Dot
Open its profile and use the available pause control if you need to stop its activity. If you are considering resetting or deleting it, read the confirmation carefully. Resetting can delete the Dot and its associated conversations, saved memories, and scheduled tasks.
OpenAI Dots Safety Tips
The convenience of an always-on agent also means its access and ongoing responsibilities deserve regular review.
Before expanding a Dot's role:
- Connect only the apps needed for the task.
- Review permissions whenever you add an integration.
- Require approval for consequential external actions.
- Keep sensitive information out of tasks that do not need it.
- Check outputs before publishing, sending, or using them for decisions.
- Review scheduled tasks periodically to ensure they are still needed.
- Pause the Dot if its activity is unexpected.
Start with a task that reads information and produces a draft or report. Once you have checked the output and understood the permissions involved, you can decide whether a more autonomous workflow is appropriate.
FAQs
How do I use OpenAI Dots?
Create a Dot in ChatGPT on desktop, connect the supported apps needed for your task, and assign a clear goal with an expected output. Configure its available approval rules, then review the task and its result.
Can I create an OpenAI Dot on my phone?
Yes, Dot creation is supported in the latest ChatGPT mobile app for iOS and Android, as well as the desktop app and desktop web. Mobile web is not supported. Some messaging-channel setup may require desktop.
Which plans support OpenAI Dots?
Access depends on the current rollout, plan, region, and workspace configuration. Pro users in supported markets and Business Premium users in supported regions are included; some managed workspaces may have access through an administrator-enabled beta. Check the official Help Center for the latest requirements.
Do I need to connect apps before assigning a task?
Not for every task. A task may not need an external app, but work involving email, calendars, messaging platforms, or other services requires the appropriate supported connection and permissions.
Can I let a Dot send messages automatically?
The available custom rules and app permissions determine what it can do. For customer-facing messages or other consequential actions, require approval until you have a clear reason and a suitable control process for allowing more autonomy.
How do I know whether a Dot completed its task correctly?
Open the Dot's profile, inspect the task status and result, and compare important claims with the original sources. A completed status does not guarantee that the output is accurate.
Conclusion
The best way to get started with OpenAI Dots is to create one for a single, clearly defined responsibility. Connect only the information it needs, specify the output you expect, and decide which actions require your approval before the task begins.
Once the first result is useful and verified, refine the instructions and consider whether the task should recur. That gradual approach is more reliable than connecting every app and delegating a broad responsibility immediately.
For wider context on the product ecosystem, read ToolJunction's overview of OpenAI DevDay 2026 announcements.
