ChatGPT Enterprise Connectors: What They Cost and Cover

Connectors are the part of an AI rollout that nobody budgets for and everybody underestimates. The seat price is a spreadsheet exercise. Getting four hundred people to individually authorize six applications is a change management project, and it is the reason a lot of ChatGPT deployments look excellent in the pilot and thin out by the second quarter.

This is the coverage half of the buying decision we price out on ChatGPT Enterprise vs Copilot pricing. That page settles what each vendor charges. This one settles what the connectors actually reach, who has to do what to make them reach it, and which of those steps an administrator can do on everyone's behalf.

How much do ChatGPT Enterprise connectors cost?

Nothing on their own. OpenAI does not sell connectors, apps or the Company Knowledge plugin as a priced add-on. They are included with paid ChatGPT Business, Enterprise and Edu seats, so the connector cost is the seat cost. ChatGPT Business Standard seats are $20.00 per user per month billed annually, $25.00 monthly. Premium seats are $100.00 annually, $125.00 monthly. ChatGPT Enterprise publishes no price at all and is quote-only on an annual commitment.

There is one genuine consumption cost hiding behind that. Heavy retrieval, deep research and agent work draw on workspace credits once a seat exceeds its included limits, and on Enterprise the credit pool sits at the contract level rather than per person. So a company that connects every app and then actually uses them can push a bill that a company with the same seat count and no connectors never sees. Budget the seats, then budget a credit buffer for the teams that will lean on it.

Which apps can ChatGPT Enterprise connect to?

The supported set covers Slack, Google Drive, SharePoint, GitHub, HubSpot, OneDrive, Box and Dropbox, alongside custom apps a workspace builds with MCP. That is a wider net than Microsoft 365 Copilot casts, since Copilot grounds its answers in the Microsoft Graph and does not natively reach Slack. It is narrower than it looks, though, once you read what OpenAI says availability actually depends on.

Verbatim from the help center: app availability depends on the app, your ChatGPT plan, region, workspace, role, model, and the ChatGPT interface you use. That is seven variables, and a buyer evaluating on a laptop in a pilot workspace is testing exactly one combination of them. Two people on the same contract can genuinely see different app sets.

Capabilities differ per app too, and this is the detail most write-ups skip. OpenAI separates them into interactive experiences, search, deep research, sync and write actions. Only some apps support sync, which it describes as indexing content in advance for faster responses and improved quality. Every other app is queried live at the moment you ask. If your Confluence connector does not sync, its content is not sitting in an index waiting to be found, and recall behaves accordingly.

Do admins have to connect apps for each user in ChatGPT?

Admins cannot connect most apps on a user's behalf. OpenAI states it twice and in almost the same words: installing the plugin does not connect your app accounts, and plugin installation does not connect individual app accounts or grant access to source data. Members authenticate a provider account when the selected app requires individual authorization. A supported administrator-managed Google Drive source is the one exception the documentation names.

Work that through for a real deployment. Four hundred people times five applications is two thousand OAuth flows, none of which an administrator can complete and every one of which is a place someone quietly stops. Nobody refuses. They just do three of the five and never come back to the others, and six months later the company has an assistant with patchy, invisible, per-person coverage that no dashboard fully describes.

This is a design choice rather than a defect, and it has a real security virtue: ChatGPT never holds access that a person does not already hold. But it is the structural reason a connected assistant is not the same product as a company index, and it is worth saying out loud before you size a contract around it.

Does ChatGPT Enterprise respect our existing permissions?

Yes, and this deserves crediting rather than attacking. OpenAI states that Company Knowledge respects permissions in the connected source and that a member can retrieve only information they are already allowed to access. The plugin does not override workspace access controls or app permissions. Enterprise and Edu administrators can manage app access with RBAC and group-level permissions, and can require SSO and SCIM for provisioning.

The permission model is genuinely sound, and any comparison that implies otherwise is selling you something. The honest critique is not that ChatGPT sees too much. It is that an administrator has no single answer to the question of what the company can currently find, because the answer is the union of several hundred individual authorization decisions.

What permissions do ChatGPT apps have by default?

The default setting is Important actions, which lets ChatGPT read from connected apps automatically while asking before anything with meaningful outside effect. OpenAI defines an important action as one that could have a meaningful effect outside ChatGPT, expose sensitive information, or be difficult to undo: sending an email, deleting content, making a purchase, moving a file, or changing sharing permissions or security settings.

The available options run Always ask, Allow read actions, Allow low-risk actions, and Allow all actions. That last one carries a useful guardrail: OpenAI notes the standard account-wide and workspace-wide permission selectors do not offer it, so a blanket allow-everything cannot be set globally in the normal path. Some especially risky actions are blocked rather than presented for approval, and a less restrictive setting does not override workspace protections.

Two defaults belong in your security review. Apps are enabled by default in ChatGPT Business, while new Enterprise and Edu workspaces start with a selected set of enabled apps. And OpenAI is explicit that ChatGPT may use the Company Knowledge plugin without an explicit mention, and may use apps automatically when Company Knowledge is not selected at all. Retrieval is not something a user has to opt into per question.

What this means when you are actually signing

If you are the person building the case, three numbers matter more than the seat rate. How many applications hold knowledge people search for. How many of those are in the supported set. And how many employees you can realistically get through an individual OAuth flow for each one. That third number is the honest denominator of the whole investment, and it is the one no vendor page will give you.

It is also worth pinning down before the paperwork moves, because a ChatGPT Enterprise agreement is an annual commitment with no public price to anchor against, and unused seats are not refundable once a workspace has committed to the seat count for that billing cycle. By the time you route the order form for signature, the seat count is effectively fixed for the year, so the coverage question needs answering in the evaluation and not after it.

Where a search layer does something different

None of this makes ChatGPT the wrong purchase. It makes it a particular kind of purchase: an outstanding assistant that borrows each person's access, one person at a time. If what your company keeps failing at is drafting, analysis and reasoning, buy it and do not overthink the connectors.

If the recurring complaint is that the answer already existed and nobody found it, the shape of the fix is different. Unified search indexes Google Drive, Slack, Notion, Confluence, Gmail, Jira and Salesforce together, returns a written answer with inline citations back to the source, and enforces each system's own access controls at query time, so a company-wide rollout never surfaces something a person could not already open. One connection per source, set up once by an administrator, rather than one per person per app. InSearch is $20.00 per user per month on Team billed annually, and the access model is covered in more depth on permission-aware AI search.

Plenty of companies end up running both, and that is a sensible outcome rather than a hedge. The mistake is assuming that connecting an assistant to eight applications is the same thing as being able to search them.