Business workflow guide
ChatGPT dots in Slack and Teams: set clear messaging boundaries
AI-assisted guidance prepared for Blue Canvas from the official sources linked below. The examples are proposed tests, not reports of firsthand product testing.
Connecting a messaging channel gives you another way to reach your dot. Before using it for team work, decide who should receive its replies, which information may be shared and what should stay in a private conversation.
What the official messaging guidance confirms
OpenAI documents Slack and Microsoft Teams as contact methods. In your dot's profile, select Add to see the methods available to you. For Slack, the guidance describes direct messages and mentions in a channel where the dot has been added. It responds to you by default; you can instruct it to engage with others.
The same documentation does not describe every Slack interaction as a Teams feature. Check the options actually offered in your account and workspace. This guide does not promise that an installation, administrator approval or particular channel action is available to everyone.
Product source: OpenAI's official documentation, checked 16 October 2026.
Write a destination rule before a messaging rule
Start with the audience. A private request about a project may contain background that is inappropriate for its shared channel. Specify the exact destination and the information allowed there, not merely 'tell the team'. Use an approved test channel or a private conversation for the first exercise.
The following is our proposed routing plan for a fictional internal project. It is not a description of settings already saved in your account. Adapt the destinations and get the relevant owner's agreement before testing.
| Update | Proposed destination | Boundary |
|---|---|---|
| Routine progress | The existing private task conversation | Keep source detail out of a shared channel |
| A decision for the project owner | The owner's agreed private contact method | Ask one specific question with its source |
| A team status update | A draft for the owner to review | Do not post until that message and destination are approved |
| Missing source access | The task owner | Report the missing source, not an assumed all-clear |
Do not mistake continuity for a shared transcript
OpenAI says these contact methods reach the same dot, while each channel retains its own visible conversation. Relevant context can carry across, but messages are not all mirrored and access to private context is not permission to disclose it.
In practice, put the essential task reference in a handover message. If a colleague needs to review an answer, include an approved source link they can open rather than relying on a private conversation they have not seen. Remove information that is unnecessary for their decision.
Product source: OpenAI's official documentation, checked 16 October 2026.
Try a draft-only channel brief
Review the draft for audience as well as accuracy. Does it reveal a private negotiation, name somebody who need not be included, or present an unanswered question as a commitment? Rewrite the brief where necessary. A polished summary can still be the wrong message for that channel.
If you later decide to send an update, review the exact wording and destination at that point. Keep a separate record of what was drafted, what was authorised and what was actually posted. That distinction makes a failed or duplicated delivery easier to investigate.
Adapt this example to your approved task
For [fictional project], use only [approved test thread] to prepare a status update. Keep the draft in this private conversation. Include confirmed decisions, open questions and a link to each source message. Do not post to the project channel or contact colleagues. Do not include information from other private conversations. Ask if the intended audience or source access is unclear.
A connected channel is not a monitoring service
OpenAI says adding a dot to a Slack channel does not create a monitoring schedule. If ongoing checks are wanted, specify the source, event or timing and notification conditions, then ask what was actually set up.
For a one-off test, do not add recurring work at all. Check the response in the intended destination, confirm that the source links work for the reviewer and inspect whether anything was posted elsewhere. Treat this as an acceptance test you perform, not an outcome guaranteed by a connection screen.
Product source: OpenAI's official documentation, checked 16 October 2026.