A product manager spends twenty minutes inside a ChatGPT Project, uploading a spec document and explaining the quirks of a vendor contract so she can get help drafting an email. Two days later, stuck on a related question, she opens Claude instead because ChatGPT is slow that afternoon. Claude has no idea any of that conversation happened. She retypes the vendor's name, the contract clause, and the part of the spec that actually matters, from scratch. That is not a bug in either product. It is the default behavior of how ChatGPT Projects, Google's Gemini, and Claude Projects are each built, and the exact line each one draws around saved context is worth knowing before a workflow gets built on top of it.
OpenAI, Google, and Anthropic each shipped a version of a saved workspace that remembers a project's files and instructions. None of the three shipped a way to move what that workspace knows into a competitor's product. That gap is not a missing feature waiting to be filed as a bug. It is the current shape of the market.
What a ChatGPT Project Actually Keeps
A ChatGPT Project is a workspace that groups related chats, uploaded files, and custom instructions so the model can draw on all three inside that one project. Project instructions apply only within that project and override whatever general custom instructions are set for the account. The detail that decides whether a project behaves like a sealed room comes down to a single setting: default memory or project-only memory.
- Project-only memory: chats inside the project cannot reference conversations outside it, and chats outside the project cannot reference anything said inside it.
- Default memory, which most plans use unless it is changed: chats can reference both project and non-project conversations, pulling in context from elsewhere in the account.
- Enterprise and Edu accounts default to the contained behavior, where project chats stay inside the project and cannot reference outside conversations.
- A shared project is automatically locked to project-only memory and cannot be switched back to default, regardless of which plan owns it.
Sharing itself is wide open, by OpenAI's own account. Project sharing works on every tier, Free, Go, Plus, Pro, Business, Enterprise, and Edu, and an owner can invite people individually, invite a group, or hand out a link set to either "only those invited" or "anyone with a link." Invited collaborators get one of two roles: chat access, which lets them read and use a project's chats, files, and instructions without inviting anyone else, or edit access, which lets them change instructions and files and add new collaborators. File limits still scale by plan, running from five files on Free up to forty on Pro, Business, Enterprise, and Edu.
Gemini's Equivalent Is Mid-Replacement
Gemini never had a feature named Projects. Its closest analog, Gems, is a saved, customizable version of Gemini built around a name, written instructions, and optional knowledge files, meant for a specific recurring task. As of this comparison, Gems are in the middle of being retired. Google announced on September 30, 2026 that a new feature called Skills will replace Gems as "the tool for tailoring instructions for specific tasks," rolling out to the Gemini app globally starting that day.
A skill is called up by typing a forward slash and the skill's name in the prompt bar, or Gemini can recognize on its own when an existing skill fits the current prompt. Google's own timeline states that Gems on personal Google accounts go away starting in November 2026 and convert automatically into skills, while Workspace business, enterprise, and nonprofit accounts keep their Gems until March 2027, and education accounts until June 2027. Gems built through Google Labs are the one exception: they are retired without converting into anything.
Sharing is where the migration shows its seams. Google's own Gems help page confirms that a Gem, or a conversation built on one, can be shared today, though a Gem someone shares with you does not appear in your own Gem manager until you open it. Skills do not have that same option yet: Google's announcement says the ability to share a skill is coming "in the coming weeks," not at launch. Google's Gems help page also does not state whether the memory feature that lets Gemini recall details from past chats applies while a Gem, or now a skill, is active. That detail simply is not addressed on the page, which makes Gemini's version of project-scoped memory the least clearly specified of the three.
What a Claude Project Actually Keeps
A Claude Project groups chats around shared project knowledge, uploaded files, and instructions that apply to every chat started inside it. That knowledge base stays inside the project. What sets Claude apart is how it splits memory into two separate layers: a project's own memory, and the account-wide memory built from everything that happens outside projects.
- Inside a project, Claude draws on that project's own memory plus account memory built from outside the project, but never on memory saved inside a different project.
- Memory is saved to a project as you chat, and Anthropic states that, if account memory is turned on, Claude can reference what it learned in a project inside chats outside of projects too.
- Turning off "Use account memory" on a project's settings page is what actually seals it off. Without that toggle, a project is not automatically a closed room.
- Memory is on by default for Free, Pro, and Max users. Team and Enterprise owners decide whether to enable it at the organization level, and it stays off for each member until they turn it on.
Sharing a Claude Project is narrower than ChatGPT's version. Anthropic's own tutorial describes sharing only for Team and Enterprise plans, where members of the same organization can share a project with each other unless an admin has turned that off. Pro and Max users get Projects, but the public documentation does not describe a way for them to share one with another person.
The One Real Crack in the Wall
Of the three, Claude is the only one with an officially documented way to move memory in or out of the product at all, and it is worth describing exactly, because it is more limited than it sounds. Anthropic's support page walks through importing memory from "your current AI service" using a prompt Claude provides, then pasting whatever that other tool exports into Claude's import box, where Claude turns the text into individual memory entries a person can review before keeping. Exporting works the same way in reverse: ask Claude to write out your memories, or view them in settings, then copy that text into a file or paste it somewhere else, including "another AI service."
Anthropic is upfront about the limits. The page calls memory imports "experimental and still in active development," warns that "Claude may not always successfully incorporate imported memories," and notes that because Claude's memory is designed to focus on work-related topics, it "may not retain imported personal details unrelated to work." There is no one-click sync in either direction, no shared standard the way email providers agree on a common protocol, and no equivalent feature announced by OpenAI or Google. A manual copy-paste job, with an explicit warning that it might drop half of what came with it, is currently the most interoperable memory feature on the market.
Three Workspaces, Side by Side
| What to check | ChatGPT Projects | Gemini (Gems to Skills) | Claude Projects |
|---|---|---|---|
| Does it have its own scoped memory | Yes, as project-only or default memory, chosen per project | Not clearly documented. Gems and skills are reusable instructions, not a described memory store | Yes, each project keeps memory separate from other projects |
| Can that context leak outside the feature | Yes by default on most plans, unless the project is set to project-only | Not addressed on Google's own Gems help page | Yes, if account memory stays on for that project; a toggle turns it off |
| Can you share it with another person | Yes, on every plan from Free through Edu, with chat or edit roles | Yes for Gems today; skill sharing is planned but not live as of this comparison | Only on Team and Enterprise plans, and only if an admin allows it |
| Can the saved context move to a rival AI product | No built-in export to another provider | No built-in export to another provider | Partially: memory can be copied out as text and pasted elsewhere manually, which Anthropic calls experimental |
| What happens when the feature itself changes | Project settings and files persist unless deleted | Existing Gems convert automatically into skills starting November 2026 for personal accounts | No announced retirement; projects and their memory persist as designed |
Why the Same Story Repeats Three Times
Line the three up and a pattern holds no matter how each company phrases its own version. A saved workspace, whether it is called a Project, a Gem, or a skill, is built to make one product better at a recurring task inside that product. None of the three documented an automatic way to hand that saved context to a rival app. Claude's copy-paste export is the single, narrow exception, and even Anthropic frames it as experimental rather than a finished feature. That pattern does not require bad intent to explain it. Keeping a person's accumulated context inside one app is simply what each company's product is built to do, and nothing about switching which AI tool someone prefers currently obligates any of them to make that easy.
The actual cost of that design lands on whoever is doing the switching, not on any of the three companies. A freelancer who built out a ChatGPT Project for one client, then opens Claude because a teammate prefers it, is not missing a hidden setting. They are hitting the exact boundary described above: project knowledge and project memory both stay where they were created, and only Claude offers even a manual way to drag some of it along. MemX (memx.app) exists for that specific gap. It is a memory layer that sits outside all three products and recalls context across ChatGPT, Claude, Gemini, and a person's own documents, so the vendor contract explained inside a ChatGPT Project is also there the next time the same question comes up in Claude. It does not take actions, browse on its own, or touch what OpenAI, Google, or Anthropic store inside their own Projects or Gems. It is private by architecture, built on per-user isolation and encryption at rest, and its only job is handing saved context back to the person who saved it, regardless of which company's chat window they happen to be typing into that day.
Before building a workflow around any Project, Gem, or skill, open that product's own current help page and check two things: whether its memory can leak outside the feature by default, and whether anything resembling an export exists. Both answers change faster than most third-party guides keep up with.
01Does ChatGPT Projects memory carry over to regular chats?
By default on most plans, yes. ChatGPT Projects can use default memory, letting chats reference both project and non-project conversations, or project-only memory, which seals the project off completely. Shared projects are locked to project-only automatically.
02What happened to Gemini's Gems?
Google is replacing Gems with a new feature called Skills, announced September 30, 2026. Gems on personal accounts are retired starting November 2026 and convert automatically into skills; Workspace business accounts keep Gems until March 2027, education accounts until June 2027.
03Can you share a Claude Project with someone else?
Only on Team and Enterprise plans, and only if an admin has not disabled project sharing. Anthropic's own documentation does not describe a sharing option for Claude Pro or Max users.
04Does Claude's memory actually transfer to ChatGPT or Gemini?
Partially, and manually. Claude lets a person copy memory out as text to paste into another AI tool, but Anthropic calls the import side experimental and warns it may not successfully incorporate everything brought in.
05Why doesn't my AI remember things when I switch apps?
Because none of the three major providers built an automatic way to move saved context between products. Each one's Project, Gem, or skill is designed to improve that single app, not to hand context to a competitor.
