The short answer
If your team works in one AI tool, use what that tool offers. In ChatGPT that is company knowledge. In Claude that is a Project.
If your team works in both, neither is enough by itself. What one holds, the other does not read. Then the knowledge has to live in one source outside both, and both tools connect to it. The home page answers the same question in three lines, in its FAQ.
What follows about ChatGPT and Claude is what OpenAI and Anthropic say on their own help pages, as we read them on 1 October 2026. Both products change often. Their own pages are leading, and each section links the one it rests on.
What ChatGPT company knowledge is
OpenAI's help centre describes company knowledge as a plugin in ChatGPT. It answers questions about your organisation from the knowledge sources that are available to you there. It replaces the earlier company knowledge option in a chat.
- Sources. It works with the supported sources available in ChatGPT, including connected apps and custom apps built with MCP.
- Plans. ChatGPT Business, Enterprise and Edu.
- What it takes. The plugin has to be available and installed, and you need at least one supported knowledge source. Installing the plugin does not by itself give access to a source's data.
- Using it. You select it in a chat with an @ mention or from the tools menu. ChatGPT may also use it without being asked.
- Permissions. It respects the permissions of the connected source. A member retrieves only what they may already access there.
- Checking. When an answer includes citations or source links, you can use them to verify it.
- Training. OpenAI states that it does not use Business, Enterprise or Edu workspace data to train its models by default.
The rest is on OpenAI's page about company knowledge in ChatGPT.
What Claude Projects are
Anthropic's help centre describes Projects as self-contained workspaces in Claude, each with its own chat histories and its own knowledge base. In a project you upload documents, give Claude the context of the work, and have chats that stay with that subject.
- Sharing. On the Team and Enterprise plans, a project can be shared with other members of the organisation.
- Can view. A member sees the project's contents, knowledge and instructions and can chat within it, but cannot edit it.
- Can edit. A member can change the project's instructions and knowledge, and add or remove members.
The rest, such as how much a project can hold, is on Anthropic's page about projects.
Side by side
Where a cell says "see the vendor's page", we did not establish the answer from the vendor's own page, and we do not answer for them.
| ChatGPT company knowledge | Claude Projects | One source for both (Dienox) | |
|---|---|---|---|
| Where it lives | Inside ChatGPT, as a plugin. | Inside Claude, as a workspace per project. | Outside both. Each tool connects to it. |
| What it reads | The knowledge sources available to you in ChatGPT, including connected apps and custom apps built with MCP. | The documents and context added to that project. | One knowledge base of how the organisation works: who decides what, what was decided, what you sell and how the work runs. |
| Which plans | Business, Enterprise and Edu. | Sharing is for Team and Enterprise. For the rest, see the vendor's page. | Dienox is in beta, in a Self-serve and an Enterprise edition. |
| Sharing in a team | Each member retrieves only what they may already access in the connected source. | Shared with members as Can view or Can edit. | Roles per organisation: administrator, editor or reader. |
| Read by the other tool | No. It is a feature of ChatGPT. | No. It is a feature of Claude. | Yes. Claude, ChatGPT and any other tool that speaks MCP. |
| Who keeps it current | See the vendor's page. | Members who can edit the project. | Everyone, as they work. A connected tool writes back what was decided, and a fact is corrected once. |
| History of changes | See the vendor's page. | See the vendor's page. | Every change is kept, with who made it and when, and can be undone. |
| Can an agent act on it, and within which limits | See the vendor's page. | See the vendor's page. | Yes. Every action is read, write or irreversible, and switched on or off per agent. An irreversible action waits for a person. |
Where each one stops
Nothing here is a fault of either product. Each does its work inside its own tool. The edge only shows when a team uses more than one.
- Two tools, two pictures. Half the team asks ChatGPT, and the other half asks Claude. What one half gave its tool, the other half's tool does not have.
- A fact is corrected in more than one place. When a decision changes, every place that holds the old one has to change with it.
- A change of tool starts from zero. What was collected inside one tool is not what the next tool reads.
A team feels this as answers that differ. Two colleagues ask the same question, each in their own tool, and get two versions of the organisation. The memory each tool keeps about one person has the same edge: AI memory for a whole team.
ChatGPT Projects, custom GPTs and Skills
People also compare Claude Projects with ChatGPT Projects, Claude Projects with custom GPTs, and Claude Skills with Projects. We did not check those features for this page, so we do not describe them here. For ChatGPT Projects and custom GPTs, see OpenAI's help centre. For Skills, see Claude's help centre. Ask of each what this page asks: which of your tools can read it?
One source for both
Both tools can read a source that sits outside them. An outside source reaches Claude as a custom connector and ChatGPT as a custom app. Both speak MCP, the open standard for connecting AI tools to data and systems.
Dienox is such a source: one knowledge base of how your organisation works, read by Claude, ChatGPT, Copilot, Cursor and any other MCP client. Dienox calls that universal AI context.
- Roles. Administrator, editor or reader. Roles decide who may change what.
- A history. Every change is kept, with who made it and when, and can be undone.
- A source under each answer. Anyone can check where an answer came from.
- A limit on every action. Read, write or irreversible, switched on or off per agent.
One thing we do not claim. OpenAI's page says company knowledge works with supported sources, including custom apps built with MCP. We have not tested Dienox as a source for that plugin. What we describe is Dienox as a custom app in ChatGPT and as a custom connector in Claude. The steps for each are on connect your AI tools to your company knowledge.
Which to pick
- One tool, and its own feature covers what you need. Use that feature. It may be all you need.
- Two or more tools. One source outside them, so every tool starts from the same facts.
- A team that may change tools. One source outside them, so your knowledge outlives the change.
- Agents that act for you. One source with a limit on every action, and a person who approves what cannot be taken back.
The two do not exclude each other. A project, or company knowledge, serves the work that lives in one tool. One source serves what the whole organisation needs to know, in every tool.
Questions people ask
Can ChatGPT read a Claude Project?
No. A Claude Project is a workspace inside Claude, with its own chats, knowledge and instructions. On Claude's Team and Enterprise plans it can be shared with colleagues, who then use it in Claude. ChatGPT does not read it. For both tools to start from the same facts, the facts have to live in a source that both connect to.
Can Claude read ChatGPT company knowledge?
No. Company knowledge is a plugin in ChatGPT, and it answers from the sources available to a person there. Claude does not read it. Claude can connect to an outside source through a custom connector, and ChatGPT can reach the same source as a custom app. That shared source is what both then read.
Which ChatGPT plans include company knowledge?
At the time of writing, OpenAI's help centre names ChatGPT Business, Enterprise and Edu. The plugin has to be available and installed, and a person needs at least one supported knowledge source. We last checked this on 1 October 2026. Plans change, so OpenAI's own page is leading.
Can a Claude Project be shared with a team?
Yes, on Claude's Team and Enterprise plans. Anthropic's help centre describes two levels. Can view lets a member see the project's contents and chat within it. Can edit lets a member change its instructions and knowledge and add or remove members. Sharing a project does not make it readable from another AI tool.
Claude Projects or ChatGPT Projects: which one for a team?
For a team, the question is who can share what. Anthropic's help centre says a Claude Project can be shared on the Team and Enterprise plans, as Can view or Can edit. For how ChatGPT Projects are shared, see OpenAI's page: we did not check it. A team that works in both tools needs one source that both read.
What is the difference between a custom GPT and a Claude Project?
A Claude Project is a workspace in Claude with its own chats, knowledge and instructions, which Team and Enterprise plans can share. A custom GPT is a feature of ChatGPT that OpenAI describes in its help centre, and we did not check it for this page. Of either, ask the same question: which of your tools can read it?
Can ChatGPT and Claude use the same knowledge base?
Yes, when the knowledge base sits outside both. An outside source reaches Claude as a custom connector and ChatGPT as a custom app. Dienox is such a source: one knowledge base that both read, with roles, the history of every change and a source under each answer. What was decided in one tool is there in the other.
Do we need Dienox if we only use one AI tool?
Not necessarily. If one tool is all your team uses, and that tool's own feature covers what you need, that feature may be enough. Dienox matters when more than one tool is in use, when your knowledge must outlive a change of tool, or when agents must act within limits you set.