/ THE SHORT ANSWER
- 01Check your plan and region before trying to set up a dot.
- 02Begin with a read-first responsibility and add app access only when the task needs it.
- 03Custom rules control actions but do not grant app access or override built-in safeguards.
- 04A two-week review can reveal whether the agent saves time or merely shifts work into checking it.
/ dotSuper point of view
A good first dot is a small operating agreement: specific inputs, a defined output, a human owner and explicit limits.
Step 1: Check access before designing around it
As of 1 October 2026, the guidance lists Pro in eligible markets, Business Premium in supported regions, and an administrator-enabled Enterprise beta.
It does not list Plus.
If the feature is absent, do not buy a new plan until you have checked the current availability and allowance for your region and workspace.
Eligible users create the dot in the desktop ChatGPT app or desktop browser.
Setup offers app connections, and the desktop app separately offers a choice to connect your own computer.
The dot has a cloud computer even if you do not connect a local one.
Step 2: Give it a responsibility, not a vague role
Start with a repeatable output and an identifiable reviewer.
For example: 'Every Friday, review the supplier update documents I provide, draft a one-page list of changed delivery dates and unanswered questions, cite each source, and ask me before sending any message.'
That instruction names the cadence, source boundary, output and approval point.
Add a date range or end condition so old recurring work does not continue indefinitely.
Tell the dot where routine updates should go and what should interrupt you immediately.
- Task: What exact decision or update should it support?
- Sources: Which apps or documents are allowed?
- Output: What format and citations are required?
- Owner: Who accepts or rejects the result?
- Boundary: Which actions require approval or are off limits?
Step 3: Connect the minimum, then set the boundary
A Slack or Teams contact method is not the same thing as permission to read a connected inbox.
Likewise, connecting your own computer is separate from the dot's cloud computer and requires your device to be online with the app open when used.
OpenAI's controls let you set custom rules for particular actions, such as asking before sending a customer message or handing deletion of shared files back to a person.
These rules supplement existing app permissions and built-in checks; they do not create a permission that an app has not granted, and the dot can still make mistakes.
For a first pilot, keep external sends, purchases, deletion, changes to customer records and publication behind human review.
The dot can draft and recommend while a named person makes the consequential decision.
Step 4: Run a two-week acceptance test
Check whether every important claim points to a real source, whether it misses changes, and whether the reviewer spends less time than before.
Only then move to fresh live work.
Open the Activity view to inspect work and requests for approval.
Review Scheduled separately for recurring work.
Pausing the dot does not automatically stop delegated tasks or cancel future schedules, so assign one person to maintain the task inventory.
| Measure | Record each week | Decision trigger |
|---|---|---|
| Accepted outputs | Number accepted without material correction | Low acceptance means narrow the task |
| Source errors | Missing, stale or invented references | A consequential error means stop and investigate |
| Review time | Minutes spent checking and fixing | If it rises, the automation is not saving effort |
| Action requests | Approvals, sends and changes proposed | Unexpected actions mean tighten permissions |
Step 5: Expand only after the work earns it
Add one new integration at a time, document what it grants, and keep the old manual route available.
If the dot needs frequent corrections, cannot cite its sources or creates more approval noise than value, pause or stop the responsibility.
That is a successful pilot outcome too: it has identified a workflow that needs cleaner inputs or a different design before automation.
What this page cannot conclude
- 01Interface steps and plan eligibility reflect OpenAI documentation accessed on 1 October 2026 and can change.
- 02The suggested two-week test is dotSuper's operating recommendation, not an OpenAI requirement or a guaranteed measure of return on investment.
- 03Custom rules and approval checks reduce risk but do not make an agent error-free.
Sources
- 01Get started with your dotChatGPT Learn · accessed Oct 1, 2026
- 02Control your dotChatGPT Learn · accessed Oct 1, 2026
- 03Getting started with your dotOpenAI Help Center · accessed Oct 1, 2026
Our editorial standard · Found an error? Send a correction with its source.
/ CITE OR SHARE THIS GUIDE
Make the evidence easy to verify.
When you reference this guide, link to its canonical URL. That gives readers one stable place for the evidence, limitations and future updates.
dotSuper Research Desk. (October 1, 2026). How to Set Up a ChatGPT Dot Without Over-Connecting Your Business. dotSuper. https://dotsuper.net/feeds/applied-systems/how-to-set-up-chatgpt-dot-safely