In a shared ChatGPT project, anyone who joins as a member can see the project's chats, files, and member list. Every member can view and download files, even with chat-only access; edit permission expands what someone can do but does not create a private folder for each person's documents. Personal memories outside the project are not exposed to colleagues. These rules come from OpenAI documentation consulted on October 24, 2026.
Who can view content and who can change it?
OpenAI separates content visibility from administrative actions. The owner controls invitations, membership, and permissions. Someone with chat access still has access to shared material, even though they cannot invite other people.
| Role | What they can access | What they can do |
|---|---|---|
| Owner | Project chats, files, instructions, and members | Create chats, manage files and instructions, invite or remove people, change permissions, and delete the project |
| Chat access | Project chats, files, instructions, and members | Chat, create new chats, and branch conversations; cannot invite people |
| Edit access | Project chats, files, instructions, and members | Can also change instructions, upload or remove files, and invite people; cannot remove existing members |
The Share panel shows invited members' names, email addresses, and groups. In managed workspaces, administrator restrictions and group permissions can also determine who has access. Chat access does not mean being limited to a single file or specific conversation.
Private invitation or open link: who can join?
The owner can choose Only those invited or, where supported, Anyone with a link. Under the second option, any signed-in ChatGPT user who obtains the address can join the project, subject to applicable limits and settings. A joining link should therefore not be treated as an invitation restricted to just one person.
In Business, Enterprise, and Edu workspaces, sharing may be limited to the workspace, and administrators may apply additional controls. Before sending a link, check its permitted audience, assigned role, and existing membership directly in the sharing panel.
What happens to files, chats, and responses?
Any member can open and download a file added to the project. Project chats are also visible to participants, including eligible conversations that a member moves from personal chat history into the shared space. The author of a chat can archive or delete it, and some conversations can be removed from the project when that option is available.
Information can also circulate in another way: ChatGPT may use shared files, instructions, and conversations as context and include their information in answers shown to other members. This does not provide automatic access to each colleague's private conversations or files from unrelated projects. It means a confidential document placed in the project may influence an answer visible to the group, even if no one opens that document manually.
Project memory is not members' personal memory
Shared projects automatically switch to project-only memory and cannot return to default memory while shared. In this mode, ChatGPT may reference conversations within the project, but not members' outside chat histories, global custom instructions, or personal memories. Chats outside the project also cannot reference its conversations through this memory mode.
This memory is a way to reuse context, not a public folder containing a list of memories colleagues can browse. Members can directly access the project's chats, files, and instructions. Memory behavior also depends on account controls and, where relevant, workspace settings.
Can the content be used to train models?
Seeing project content and using it for model training are separate issues. For personal accounts, OpenAI says shared-project data may be used to improve models only if the owner and every contributor have Improve the model for everyone enabled. People who do not want new eligible conversations to contribute can turn it off under Settings > Data controls or opt out through OpenAI's Privacy Portal.
By default, OpenAI says it does not train on content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces. Organizations still need to check their internal rules on sharing and retention. For personal accounts, the data controls documentation also notes that voluntarily providing feedback on a response can allow the associated conversation to be used for training even after opting out.
Sharing one chat is different from sharing the project
A shared link to a conversation from a private project gives access only to that conversation's content, under the link's conditions, not to the project's other chats, files, or instructions. Inviting someone to the project instead makes the shared workspace visible. If the chat is already inside a shared project, the documentation says the link recipient must join the project to view it. These options are not interchangeable.
Checklist before inviting colleagues or uploading sensitive files
- Review what is already inside: check older chats, attachments, sources, and instructions before enabling sharing.
- Separate data that does not need to circulate: confidential contracts, personal information, credentials, and third-party documents require authorization and a risk assessment before upload.
- Choose the audience: prefer Only those invited unless there is a clear reason to allow joining through a link.
- Give the minimum necessary role: chat access already allows file viewing and downloading; reserve editing for people who need to change project materials or invite collaborators.
- Check memory and data controls: project-only memory limits the context used; training controls govern a different purpose and must be reviewed separately.
- Review members after inviting: the Share panel displays participants, and the owner can adjust permissions or remove people.
What if the project was shared by mistake?
The owner can remove members and switch the link setting back to Only those invited. In managed workspaces, administrators can impose further restrictions. This limits future access but cannot retrieve files someone already downloaded or information they copied. Deleting the project is a broader, irreversible action subject to retention rules and copies held elsewhere.
The practical test is to assess the sensitivity of the entire workspace, not just the file that prompted an invitation. This is documentation-based guidance drawn from OpenAI's official pages, not the result of a DTech permissions test or a submitted project.
