ChatGPT Custom GPTs vs Claude Projects for Sharing Team Knowledge

If your team needs one place to keep shared documents and standing instructions for an AI assistant, Claude Projects is the safer choice today, because Projects are a current, documented feature with sharing roles for Team and Enterprise plans. ChatGPT Custom GPTs do the same job, but OpenAI has announced that they will be replaced by plugins, with retirement currently scheduled for 11 December 2026. This article compares how each works, what you can verify about limits and permissions, and how to decide.

Who this is for and what you will have

This is for team leads, operations managers and IT admins who want colleagues to ask questions against the same policies, product sheets or process documents without everyone uploading files each time. By the end you will know how each tool stores knowledge, who can share and edit, where the limits are, and which option fits your situation.

Plan names, limits and features change often in this area. Everything below that describes Claude comes from Anthropic's help center pages read on 26 October 2026. The Custom GPT retirement details come from trade press coverage of OpenAI's announcement, because OpenAI's own help pages weren't retrievable when this article was prepared; check OpenAI's pages before you decide.

What each tool is

Claude Projects

A Claude Project is a workspace in claude.ai. Anthropic's help center says files, text or code snippets you add to the project knowledge are used as context in every chat in that project. You also write project instructions with the Set project instructions option, which Claude applies to every chat in the project. Context from one chat isn't carried into others unless the information is added to the project knowledge.

Projects are open to all plans, with a limit of five projects on the free plan. Creating one is simple: open Projects from the left side of your account, select + New Project, name it, add a description, and then add knowledge and instructions.

ChatGPT Custom GPTs

A Custom GPT is a configured version of ChatGPT with its own instructions, optional knowledge files and optional actions that call external APIs. Teams use them to give staff a pre-briefed assistant, such as an HR policy helper.

The key fact for any new decision is the retirement. According to Virtualization Review's coverage of the announcement:

• OpenAI announced the change on 11 September 2026 and currently schedules Custom GPT retirement for 11 December 2026, with timing that can vary by account and workspace.
• For Enterprise workspaces, creation of new Custom GPTs ends on 26 October 2026, and Enterprise customers with approved deferrals can keep using them until 11 February 2027.
• Plugins replace Custom GPTs. A plugin can contain skills (reusable instructions and resources for repeatable workflows), an MCP server that exposes tools and external systems, or both.

OpenAI's own page for this is the Custom GPT retirement and migration FAQ on help.openai.com; read it for the current dates for your plan.

Side-by-side comparison

How knowledge works in a Claude Project

When project knowledge fits within the context window, Claude loads it as context for chats in the project. When it approaches the limit, paid plans automatically enable retrieval-augmented generation (RAG) mode. Anthropic's help center describes this as follows:

• It is available on Pro, Max, Team and Enterprise plans.
• Instead of loading all project content at once, Claude uses a project knowledge search tool to retrieve only what is relevant to each question.
• It expands project capacity by up to 10 times.
• It turns on automatically, and can switch back if knowledge drops below the threshold.
• A visual indicator shows that the project is RAG-enabled, and you may see Claude invoking the project knowledge search tool.
• You can't manually control when RAG is used, because activation is handled automatically based on the size of your project knowledge.

The practical lesson is to keep documents well structured, with clear headings and one topic per file, so retrieval finds the right section. You can't tune the retrieval, so verify answers on questions that matter.

Sharing and permissions

Claude

Sharing is only available on Team and Enterprise plans, and admins can turn it off. When you create a project on those plans you choose whether to keep it private or share it with your organisation. Members get one of two roles:

• Can view: see the project's contents, knowledge and instructions, and chat in it, but not edit.
• Can edit: change instructions and knowledge, update member settings and contribute.

Only the project creator can remove a member's access. A sensible setup is a small group of editors who maintain the knowledge and a wider group with view access, so the source documents stay consistent.

ChatGPT

Whether and how a GPT can be shared depends on plan, workspace policy and the GPT's configuration, so you need to check the workspace settings your admin has applied. For anything built now, also consider what happens at retirement. According to the trade coverage, migrated plugins start private, and existing GPT users don't automatically get access, so sharing has to be set up again.

What happens when you migrate a Custom GPT

If you already have Custom GPTs, the coverage describes this conversion path through the My GPTs interface:

OpenAI warns that a migrated plugin "may respond differently", so retest prompts, use of reference material, output formats and integrations before relying on it.

A neutral way to decide

Choose by these criteria rather than by brand.

1. Where does your team already have licences? A team already paying for ChatGPT business plans and one already on Claude Team will each want to stay inside the tool they use. Check that your plan supports the sharing you need before you build.
2. How long must the setup last? A Custom GPT built now may need migration within weeks. If you build one anyway, keep the instructions and files in a document repository so they can move.
3. Do you need external actions? Custom Actions that call your systems don't migrate automatically. Claude Projects don't provide actions; for tool calls you would use other mechanisms such as MCP servers or an automation platform.
4. How large is the knowledge base? Small sets of documents work in either tool. Large sets depend on how retrieval behaves; Claude documents automatic retrieval mode on paid plans, and you should test your own questions in whichever tool you pick.
5. How sensitive is the content? Check the data handling terms for your plan, who can see shared content, and whether admins can restrict sharing. Don't upload regulated or confidential data until your policy approves the tool.

Setting up a shared Claude Project

1. Open Projects at claude.ai and select + New Project.
2. Name it for the topic, for example HR policies, and write a description.
3. On a Team or Enterprise plan, choose to share it with your organisation, then set members to Can view and add editors with Can edit.
4. Add documents to the project knowledge with the + button. Prefer current, final versions, and name files clearly.
5. Select Set project instructions and write rules for the assistant.
6. Test with real questions from colleagues and fix gaps in the documents or instructions.

A starting set of instructions:

You answer questions using only the documents in this project's knowledge.
If the documents don't answer the question, say so and suggest who to ask.
Quote the document name and section for every policy statement.
Do not give legal or medical advice.

Keeping the rule to answer only from the documents reduces the chance that the assistant fills gaps with guesses, but it doesn't remove it, so spot-check answers.

Maintaining shared knowledge

• Assign an owner for each project and a review date for each document.
• Remove outdated files, because both tools will happily answer from stale text.
• Keep a master copy of every document in your normal storage. The AI tool is a consumer of the documents, not their home.
• Record the instructions in a text file so they can move to another tool.
• Re-test a set of 10 to 20 standard questions after every major change.

Troubleshooting

The assistant ignores a document

In a project in retrieval mode, only relevant passages are fetched. Rephrase the question to match the document's wording, split long documents into topical files, and add headings.

Colleagues can't see the project

Sharing requires Team or Enterprise, and admins can disable it. Check your plan and the admin setting, and confirm members were added with at least Can view.

Answers differ between colleagues

Context isn't shared between chats unless it is in the project knowledge. If one person pasted extra information into a chat, others won't have it. Add shared facts to the knowledge or instructions.

Checklist

• Confirm plan and sharing support before building.
• Keep sources and instructions in your own storage.
• Use view-only access for most members and a few editors.
• Test with real questions and verify important answers.
• Don't start new long-term builds on Custom GPTs without a migration plan.

References

• How can I create and manage projects? (Claude Help Center)
• Retrieval augmented generation (RAG) for projects (Claude Help Center)
• OpenAI to Retire Custom GPTs, Replace Them with Plugins (Virtualization Review)
• Custom GPT retirement and migration FAQ (OpenAI Help Center)