ChatGPT Company Knowledge Connectors: Connect and Check Sources

ChatGPT company knowledge connectors give ChatGPT an authorized route to search and retrieve information from internal sources. Start by enabling an approved connection, authorizing the intended account, and checking its answers against documents you already know. A successful sign-in is the beginning of that check, not proof that every useful source is available.

Tools like fenbs keep the team’s decisions and current work beside that document evidence, so an answer can lead to a clear next step. Use the source system for the original material and a shared work record for what people agreed to do about it.

How do you connect your first internal source?

1. Choose one ordinary work question, such as finding the approved customer onboarding checklist. Identify the source that should answer it before connecting anything.
2. Ask your workspace administrator to confirm that the relevant plugin or app is available to you. Check the current source capabilities in your workspace rather than relying on an old connector list.
3. Follow the offered authorization flow with the intended work account. Read the requested access and confirm which organization or workspace you are connecting.
4. Use the connection in the intended ChatGPT experience. Ask for one known document, then a question whose answer requires reading that document.
5. Open the returned source and compare the answer with it. Record missing material, incorrect interpretation, and any unexpected access before expanding the setup.

OpenAI’s plugin controls documentation↗ (opens in a new tab) separates plugin availability, connection capabilities, and authorization in the source service. Enabling an integration does not itself give it permission to read the company’s files. Our guide to ChatGPT connectors and apps explains the surrounding terminology.

Which connections can supply company knowledge?

Check what the particular integration can do. Searching for matching material, fetching a complete source, and changing a record are different capabilities. A connection that can create a document is not automatically a useful source for answering questions about existing documents.

For custom sources, OpenAI documents company knowledge compatibility↗ (opens in a new tab) through read-only MCP tools. MCP is a standard way for an AI app to use an outside tool. The guidance describes standard search and fetch tools and source URLs that people can open; your technical team should check those requirements before promising a custom source will work.

Start with a small collection whose owner can answer questions about it. For example, use the approved onboarding folder rather than every document containing the word onboarding. Specific integrations have their own setup details; the Confluence connection guide covers that neighboring case.

Whose permissions does a connection use?

Identify the account behind the connection before judging its access. With a personal work connection, inspect that authorized account. With a company-managed connection, inspect the designated account and the people allowed to use it. These may be different identities with different access.

OpenAI’s workspace connections guide↗ (opens in a new tab) explains that eligible shared workflows can use company-managed accounts. Such a connection may read a source that a participating person cannot open directly. A citation does not grant that person access to the original file, so review the connection audience as well as the account’s resource permissions.

• Confirm the account, organization, and resource collection in the authorization.
• Test an approved source with an ordinary intended user, not only the administrator.
• Use a harmless restricted test document to check an access boundary.
• Confirm who will own the connection when its original setup person changes roles.

How should you test a company knowledge answer?

Use questions with known answers before asking for broad advice. Test both finding a document and interpreting it correctly. If a checklist says that a manager approves an exception, an answer that merely mentions the checklist has not established who must approve the exception.

Try this request: “Using the approved onboarding checklist, explain the steps before a new customer receives access. Link the source for each requirement. Identify anything you cannot verify and distinguish mandatory steps from suggestions.” Replace the example with your real question and authorized source.

For work tracked in an issue system, an example using Linear with ChatGPT can help you frame a concrete retrieval question. Keep your acceptance test about the information you need, rather than whether the response sounds polished.

What should you check when material is missing?

First open the source with the actual connected account. Then verify that the intended integration is available and selected for the experience you are using. Ask for the exact document title or link to narrow the problem. Distinguish “the connection cannot retrieve it” from “the answer retrieved it and misunderstood it.”

Avoid fixing a missing result by granting broad access immediately. Capture the failing question, expected source, account, and visible error without copying sensitive content into a support request. The connector troubleshooting guide covers a fuller diagnosis when the first checks do not resolve the problem.

Where should the team keep decisions made after search?

Finding a document does not settle a choice. Say the onboarding checklist allows an exception, and the manager approves one for a particular project. The original checklist remains the reference. The approval, its scope, and the follow-up task need a shared home so another teammate does not ask ChatGPT to decide the exception again.

Keep those agreements in fenbs as Decisions and rules, with tasks for the remaining work and context notes for relevant background. People decide; assistants can record what they chose. The shared work workflow gives connected assistants a concise starting point, while the team brain collection explores the broader habit.

How to do this with fenbs

1. Create a free account, then add a project for the work your company knowledge question supports. One person can start free, with no credit card.
2. Add the agreed decision and its source reference, recording who decided and the scope of the agreement. Keep unresolved choices open.
3. Add follow-up tasks and write the next step. Use a team kanban board with appropriate member roles when colleagues need the same record.
4. Follow the integration guide for your AI app. Ask it to read the project context before work and save progress before stopping.

fenbs does not import ChatGPT chats or provide a search index for your company’s document library. If your only need is finding a handbook page, use the authorized document connection. Add fenbs when people and assistants need to carry the resulting decisions and work into the next session.