ChatGPT in Slack and Microsoft Teams: How @ChatGPT Works

ChatGPT in Slack and Microsoft Teams now runs through a shared workplace assistant you call with @ChatGPT. Mention it in a channel or thread, or message it directly, and it can answer questions, summarise a discussion or draft work using only the tools and permissions your admins have set up. OpenAI’s release notes date the launch to September 24, 2026 for Enterprise and September 29, 2026 for Business.

It is not your personal ChatGPT dropped into a chat window. It runs on its own service account, so it does not carry your memory, your custom instructions or your access to private files. That design choice shapes setup, data handling, cost attribution and the risks a training team should check first.

This guide covers how it behaves, which plans include it, what admins must do, how it differs from OpenAI dots, and where an L&D team might use it. It is part of our coverage of the OpenAI DevDay 2026 announcements, and everything here is current as of September 30, 2026.

What is @ChatGPT in Slack and Microsoft Teams?

@ChatGPT is a shared assistant that lives inside Slack and Microsoft Teams and works for a whole workspace, not for one person. OpenAI’s help centre says it is available to Business and Enterprise users, and users can ask questions, summarise information or draft work in a conversation without leaving the chat tool.

Thurrott’s DevDay upgrade roundup quotes OpenAI: “mention @ChatGPT in a channel, thread or DM.” Engadget adds that it can use tools connected to Slack and Teams, for example to generate slides from a conversation, in its ChatGPT Space coverage.

OpenAI describes two routes: mention @ChatGPT inside Slack or Teams, or connect those apps inside ChatGPT itself. This guide covers the first. The official overview covers both.

How does ChatGPT in Slack work in channels, DMs and threads?

You mention @ChatGPT with a request, and it replies in the same conversation. You can then follow up for revisions in the same thread. Channels need @ChatGPT to be added first, and administrators can restrict where it is available. Direct messages to @ChatGPT are a separate path and work without a channel.

Per the OpenAI help article on using @ChatGPT, replies in shared conversations are visible to everyone taking part. That matters for anything sensitive. An answer given in an open channel is public to that channel.

At launch, @ChatGPT does not record or access memory, so it will not build up knowledge of your team over weeks. OpenAI also says the rollout is gradual, so some features may not be in every workspace yet.

Does @ChatGPT use my personal ChatGPT account?

No. OpenAI states that @ChatGPT runs through a service account configured for the assistant, and direct messages to it do not switch it to your personal account. Your saved memories, custom instructions, skills and chat history do not carry over, and it does not inherit your access to files, apps or private channels.

If a request needs one of your personal apps, you review and approve access, and OpenAI says your approval does not let other people use your apps by asking @ChatGPT. Usage from that work is attributed to you. Everything else goes to the service account.

Which plans include ChatGPT in Slack and Teams?

Business and Enterprise workspaces get @ChatGPT, and OpenAI’s admin setup article also lists Edu accounts. Plus and Pro users do not get the shared assistant, though they can connect Slack to their own ChatGPT account as a separate feature. Sources differ slightly on plan lists, so check your workspace as of September 30, 2026.

Here is how the sources line up.

On cost, Thurrott says teammates can refine results without individual ChatGPT licenses, and OpenAI’s feature page frames it the same way. Neither gives a price for the assistant itself, so ask your account manager how seats and credits apply. Our ChatGPT pricing guide, covers the wider plan picture.

How do admins set up @ChatGPT in Slack and Microsoft Teams?

Admins connect the messaging platform in the OpenAI admin console under Agents, create a named “surface”, link company tools and test it in a small channel before widening access. Three roles are involved: a ChatGPT workspace admin, a Slack or Teams admin, and a connection owner for each tool. OpenAI’s help centre has the authoritative steps.

The steps below come from OpenAI’s setup guide for @ChatGPT. Labels may change, so treat that page as the source of truth.

How do you set up ChatGPT in Slack?

Step 1: Connect the Slack workspace
In the admin console, open Agents, choose Add and follow the Slack prompts. A Slack owner or app manager must complete the approval, and you may need to reinstall the app if scopes are missing.

Step 2: Create a surface and restrict channels
Name the surface and select Create surface. You can turn on “Selected channels only” and add test channels. Channel restrictions do not restrict direct messages.

Step 3: Test in a fresh thread
Add @ChatGPT to a test channel, mention it in a new thread, then test a DM separately.

How do you set up ChatGPT in Microsoft Teams?

Step 1: Connect the tenant
In the admin console, open Agents, follow the Teams prompts, sign in with an authorised Microsoft account and complete organisational consent.

Step 2: Install the app for the right people
Choose Open Teams admin center, then Users and groups and Install app. Pick the users or groups, confirm your Teams policies allow the app, and return when the connection shows “Ready”.

Step 3: Create and test the surface
Name the surface, select Create surface, then mention @ChatGPT in an approved channel and confirm it answers.

After that, add company tools from the surface’s Plugins section and write admin instructions covering tool use and company terminology. Admins can also set a default spend limit under Usage limits, with per-surface overrides.

What data can @ChatGPT see, and what should admins worry about?

@ChatGPT sees the conversation it is invited into plus whatever the connected company tools expose. It does not inherit a user’s access. The connected account’s permissions decide what it can read and do, so an over-permissioned connection is the main risk. Replies are visible to conversation participants.

OpenAI’s guide says connecting a tool does not open every company resource, because the connected account’s permissions still apply. The ChatGPT in Slack and Teams feature page adds that @ChatGPT asks for approval before acting in connected tools and that audit logs track agent activity.

For compliance questions, note a distinction. The personal Slack connector inside ChatGPT is a third-party integration: OpenAI’s Slack app article says relevant requests and data may be sent to Slack under Slack’s terms, and that ChatGPT’s data residency controls do not apply to Slack data. The pages I could read do not say whether the same applies to @ChatGPT, so ask OpenAI.

On regions, the overview says @ChatGPT is available globally to Business and Enterprise users. Dots were reported by Fortune as unavailable in the EU, Switzerland and the UK, so confirm both with your account team.

How are dots, Agent 365 and the Meetings plugin connected to @ChatGPT?

@ChatGPT is a shared team assistant. Dots are personal, always-on agents you message in ChatGPT, Slack or Teams. They share the chat surface but not the model of ownership. Only a dot’s owner can direct it, and OpenAI says it is working with Microsoft to bring specialist dots into Agent 365 governance.

This side-by-side view separates them.

The dots admin article lists a separate “Add dots to Slack and Microsoft Teams” permission and says a Slack workspace owner must approve installation if your workspace requires app approval. Enterprise model controls do not apply to dots. Our guide to OpenAI dots covers the wider feature.

On Agent 365, Benzinga’s report on Microsoft and dots describes it as a control plane using Entra for identity, Defender for security and Purview for data governance, and lists a standalone price of $15 per user per month with an annual commitment. That is Benzinga’s figure, not OpenAI’s.

The ChatGPT Meetings plugin is a separate tool. It records meetings on the Mac desktop app and saves notes into a ChatGPT Space. I found no source saying @ChatGPT reads those notes directly in Slack or Teams. One possible path, if your plan allows it, is sharing a Space with a team, as OpenAI’s September 29 release notes describe.

What are Team Tasks and how do they handle recurring work?

Team Tasks let a team delegate recurring work to ChatGPT on a schedule or trigger, using approved company tools. OpenAI’s Enterprise notes give the example of posting weekly project updates to Slack. Tasks run in the cloud on the team’s service account, not on a member’s credentials.

Workspace owners must grant “Create teams” and “Create and manage team automations” permissions, per OpenAI’s teams and Team Tasks documentation. Admins must also enable “Allow event-triggered scheduled tasks” for event triggers.

There are limits. Team membership cannot sync with workspace groups or Slack and Teams channels, and all members see previous runs and files. Team Tasks also draw on workspace credits, with separate team spending limits.

What are the differences between ChatGPT in Slack and in Teams?

Functionally, OpenAI treats the two platforms almost identically: same assistant, same service account model, same admin console. The differences sit in setup and control. Slack uses app approval and optional channel restrictions. Teams uses Microsoft consent, Teams admin center installation and user or group targeting.

If your organisation runs both, write one policy and apply it twice. Our guide to training software for remote teams explains why chat tools are now part of the training stack.

How could L&D teams use ChatGPT in Slack?

An L&D team could use @ChatGPT for three low-risk jobs: answering repeat learner questions in a support channel, summarising a long cohort thread for facilitators, and drafting programme announcements. These are possibilities, not documented case studies. Each needs human review and a clear source of truth.

Learner FAQs are the obvious start. Admin instructions can name your programmes and terminology, such as SCORM versus xAPI. Point it at an approved FAQ document that an instructional designer or SME owns.

Cohort threads are the second: a facilitator returning from leave could ask for a summary of open questions, then check it against the thread. Announcements are the third: draft a reminder or compliance-training deadline message, then edit before posting.

Here is a sample admin instruction an L&D admin might adapt.

You support our learning team. Answer only from the approved FAQ and policy documents you are connected to. If the answer is not there, say so and tag the programme owner. Use our terms: cohort, learning objective, microlearning, LMS. Never confirm grades, completion records or personal data.

Keep learner records out of scope. Completion data, Kirkpatrick evaluation results and HR information belong in your LMS or TMS, with their own permissions. Our roundup of the top AI-powered LMS platforms and the guide to GPT-6 Astra for learning and development cover where AI belongs inside the learning stack.

What should a rollout checklist for @ChatGPT in Slack and Teams include?

A good rollout checklist starts with a pilot channel, confirms permissions with real users, sets spend limits and writes usage rules before anyone else gets access. OpenAI’s own validation list is the best starting point: conversation access, source access, access limits, write actions and output access.

1. Name owners for three roles: ChatGPT workspace admin, Slack or Teams admin, and each connection owner.
2. Pick one pilot channel or team and one narrow use case.
3. Connect only the tools the use case needs, using a least-privilege account.
4. Write admin instructions with terminology, approved sources and refusal rules.
5. Run OpenAI’s validation tests, including a harmless out-of-scope request and a check that output recipients have permission to see the result.
6. Set a default spend limit and review usage weekly during the pilot.
7. Tell staff that channel replies are visible to participants and that AI answers may be inaccurate.
8. Check region and legal requirements before extending beyond the pilot.

Run dots or Team Tasks as separate pilots. They have their own permissions and are off in Enterprise until an admin enables them.

Conclusion

Your next step is small: read OpenAI’s admin setup guide for @ChatGPT, then pick one channel and one job. For a training team, that might be a learner-support channel with an approved FAQ. Run OpenAI’s validation tests before inviting anyone else.

While you pilot, remember what is still moving: plan lists differ between OpenAI pages and the assistant’s own pricing is not published in the sources I found. Our OpenAI DevDay 2026 recap tracks the wider changes, and our complete guide to AI tools helps you compare alternatives before you commit.