A field note · ~6 min read
Yes — but it is a different shape. Claude carries persistent details across conversations and gives you Projects, which are walled gardens of context you load deliberately. ChatGPT has one global memory bucket that follows you everywhere automatically. One is a filing cabinet. The other is a sticky note on the monitor.
| ChatGPT | Claude | |
|---|---|---|
| How it stores | One global memory list, applied to every chat | Persistent memory plus per-Project context |
| Who decides what's kept | Mostly the model, automatically | More of it is you, deliberately |
| Carries across chats | Yes, everywhere | Memory yes; Project context stays scoped |
| You can read what's stored | Yes — Settings → Personalization → Memory | Partly, and Project files are visible |
| You can export it | No | No |
| Survives a model change | Not guaranteed | Not guaranteed |
It depends on what you are asking it to remember.
For a single body of work — a manuscript, a legal matter, a codebase, a research thread — Claude Projects usually wins. You choose what goes in, it does not compete for space against unrelated facts about your dog, and the context is deep rather than broad.
For everyday continuity — the assistant simply knowing who you are without you setting anything up — ChatGPT feels better, because the memory is automatic and global. You never configure it. It just slowly learns a thin outline of you.
Thin is the operative word. ChatGPT's memory bucket is small — see how much ChatGPT actually remembers. Claude's Projects are deeper but do not travel between Projects, let alone between providers.
Short answer
Claude does have memory, and in some ways a better-organized one. But both Claude and ChatGPT store their version of you on their own servers, in their own format, with no export. Whichever you prefer, the thing you build up inside it is not yours to take with you.
Neither provider will hand you your memories. There is no "export memories" button in ChatGPT and no equivalent in Claude. You can read them on screen. You cannot take them.
That is fine until the day you want to switch, or a model you liked gets deprecated, or a rollout resets behavior you had grown used to. At that point two years of accumulated context becomes something you have to rebuild from scratch, by hand, in a new box.
The workaround is to keep the durable version somewhere neither provider controls, and feed it into whichever model you are using that week. That is the whole idea behind moving your ChatGPT memory to Claude — and, now that both support it, behind connecting a memory vault over MCP.
Related reading: Konshus vs Claude Projects, when ChatGPT memory is full, and the hidden cost of switching models.
Konshus holds the durable version of you outside both providers — and connects to Claude and ChatGPT alike, so switching costs you nothing.
Meet Konshus