ChatGPT Projects and Claude Projects diverge most clearly when you share one. OpenAI says every member of a shared ChatGPT project can view its chats and files. Anthropic says a shared Claude project exposes its knowledge and instructions while each person's chats stay private unless they share them separately. That visibility choice matters more than most feature-list differences.
For individual work, there is no decisive overall winner. Both products give you a workspace for related chats, reference material, and project-specific instructions, as independent coverage of ChatGPT Projects and Claude Projects also described. Choose based on the model and tools you already prefer. For teamwork, choose based on whether collaborators should see the conversation trail or only the common source material.
The short answer
- Choose ChatGPT Projects when the chat history is part of the team's work. OpenAI says members of a shared project can see and continue from its chats, files, and instructions.
- Choose Claude Projects when people need a shared reference base but private working conversations. Anthropic says project members can use the same knowledge and instructions without seeing one another's chats by default.
- Call it a tie for a private writing, research, or planning workspace. Both products cover the core job. File capacity and memory behavior differ, but the vendors do not publish directly comparable limits.
ChatGPT Projects vs Claude Projects at a glance
| Question | ChatGPT Projects | Claude Projects |
|---|---|---|
| Who can create one? | OpenAI documents Projects for free and paid ChatGPT accounts. | Anthropic documents Projects for all Claude users, with Free accounts capped at five projects. |
| What becomes shared context? | OpenAI lists project files, saved sources, project instructions, and eligible project chats. | Project knowledge and project instructions. Chat recall is handled separately through memory and chat search. |
| Do project instructions override account instructions? | Yes. OpenAI says project instructions apply only inside that project and override global custom instructions. | Project instructions apply only to chats in that project. Anthropic does not document an override order on that page. |
| Can a team share the project? | OpenAI documents individual sharing on Free, Plus, Pro, and Go accounts. | Anthropic limits project sharing to Team and Enterprise organizations. |
| Can project members see one another's chats? | OpenAI says all project members can view project chats and files. | Anthropic says no, unless a chat is shared separately. |
The last row is the practical dividing line. The rest needs more context because the products use different definitions for memory and knowledge capacity.
What persists across chats?
ChatGPT can reference other chats in the same project
ChatGPT Projects support default memory and project-only memory. With project-only memory, a chat can reference other conversations inside the project but cannot reference conversations or saved memories outside it. Shared projects switch to project-only memory automatically.
Default memory is less uniform. OpenAI documents different behavior by plan and workspace setting. On non-Enterprise accounts, a project using default memory may draw on conversations outside the project. Enterprise and Edu project chats stay contained. If separation matters, use project-only memory and confirm that the required memory settings are enabled.
Claude separates project knowledge from chat recall
Claude's durable layer is the project knowledge base. Anthropic says uploaded knowledge is available across the project's chats, while conversation context is not shared across chats unless it is added to project knowledge.
Claude also has a separate memory system. Anthropic's documentation says each project receives its own memory space and project summary, and paid plans can search previous conversations within the same project. Memory is enabled by default on individual Free, Pro, and Max accounts, while Team and Enterprise owners control whether their organizations can use it.
The operational lesson is simple: put approved facts, source documents, and standing requirements in project knowledge. Do not rely on either product to extract every durable detail from a prior conversation.
For a deeper look at Claude's setup and constraints, see how to use Claude Projects.
Which project holds more knowledge?
The vendors do not provide a clean, like-for-like answer.
OpenAI's Projects page currently lists five files per project on Free, 25 on Go and Plus, and 40 on Edu, Pro, Business, and Enterprise. It also says no more than ten files can be uploaded at once. However, OpenAI's separate File Uploads FAQ still lists 20 files per project for Plus. Because those first-party pages disagree, check the limit shown in your account before designing a workflow around it.
Anthropic describes capacity in relation to the model's context window rather than publishing a project file count. It says Claude can automatically switch project knowledge to retrieval-augmented generation as the knowledge base approaches the context limit, expanding capacity by up to ten times. That is Anthropic's product claim, not an independent benchmark.
Anthropic's own pages also disagree about plan coverage. The dedicated RAG article says the feature is available on all Claude plans, while the Projects overview says enhanced RAG knowledge is limited to paid plans. The documentation supports neither a reliable free-plan verdict nor a claim that Claude holds more than ChatGPT. A file count and a retrieval multiplier measure different things.
Choose Claude for a large reference collection only if retrieval-based knowledge is the capability you need and your account shows it as available. Choose ChatGPT when a clear per-project file allowance fits the material, but verify the live cap first.
Project instructions and model access
OpenAI and Anthropic both document instructions scoped to a project. In ChatGPT, project instructions override global custom instructions. In Claude, project instructions apply to the project's chats, but Anthropic's personalization documentation does not specify whether they override or combine with account-level instructions. If two instruction layers could conflict, keep the project version self-contained and remove duplicated rules.
If you need a reusable assistant rather than a workspace organized around related chats and files, see how to create a custom GPT.
Model access is not a strong reason to choose one project system. OpenAI says model and tool limits apply to project chats in the same way they apply to regular chats. The same page lists standard project tools and notes that some additional tools depend on the subscription.
Claude's model menu lets a user change the model during a conversation, subject to account access and Enterprise admin controls. Anthropic's Projects documentation does not publish a separate model roster for project chats. Since available models and admin policies change, compare the selectors in the accounts you will actually use instead of choosing from a static model list in an article.
Sharing is where the products make different choices
ChatGPT shared projects treat chats as team-visible project material. OpenAI offers chat and edit access. Chat access allows a member to see and interact with the project's chats, files, and instructions. Edit access also allows changes to instructions and files, plus invitations. A shared project uses project-only memory, so it cannot draw on a member's personal context outside the project.
Claude shares the workspace but keeps conversations personal. On Team and Enterprise plans, members with view access can use the project's knowledge and instructions; members with edit access can change them. Anthropic says each person's chats remain private unless that person shares a chat explicitly, even when the project itself is visible across the organization.
That produces two different collaboration patterns:
- A campaign room where drafts and decisions should be visible to everyone: ChatGPT fits the documented behavior better.
- A policy or research library that several people query independently: Claude preserves private working chats while sharing the common knowledge base.
- A solo project that may be shared later: decide before adding sensitive exploratory chats. Sharing a ChatGPT project changes who can see its chat history; sharing a Claude project does not expose those chats by default.
How to choose without guessing
Run the same small evaluation in both products before moving a large body of work.
- Add one representative source document and the exact project instructions you expect to keep.
- Start a new chat and ask for an output that depends on both the source and the instructions.
- Start another chat and ask for a detail that appeared only in the first conversation. Record whether the product recalls it, searches for it, or misses it under your current memory settings.
- If collaboration matters, share a temporary project with a colleague and confirm what each permission level reveals.
- Check the model selector and the file allowance shown in the account. Vendor documentation can lag behind the interface, as the conflicting file and RAG pages demonstrate.
Use the result to choose the workspace, then keep your durable context explicit. Files and project instructions are easier to inspect than remembered conversation details.
So, are ChatGPT Projects better than Claude Projects?
No broad ranking is supported by the documentation. ChatGPT Projects are the clearer choice when collaborators should share the conversation record. Claude Projects are the clearer choice when collaborators should share source material while keeping their own chats private.
For private individual work, the two are close enough that model preference, available tools, and the rest of your existing workflow should decide. The better project is the one whose boundaries match how you expect information to move between chats and people.