A tech headline this fall described Claude's new memory feature as something that "syncs with ChatGPT." Read quickly, that phrase suggests a live connection: tell ChatGPT something on Tuesday, and Claude somehow already knows it by Wednesday. That is not what Anthropic built. What actually shipped is a manual, one-time copy step. You ask one chatbot to write down what it remembers about you, then paste that text into the other one. There is no background connection, no ongoing update, and no two-way channel between the two products. The word "syncs" is carrying weight the feature itself does not, and the gap between the two matters most for anyone who actually plans to rely on an imported memory for something current.
Import and sync describe two different relationships. One is a single, manual handoff. The other is a standing connection that keeps updating on its own. Claude's memory feature is the first kind, not the second.
The Headline Behind the Confusion
The claim traces back to a Tom's Guide article covering Claude's memory launch, published under a headline built around the idea that Claude's memory "syncs with ChatGPT." The article itself walks through a real feature: Claude's ability to accept a pasted memory export from another AI assistant. But the headline's verb choice implies an ongoing relationship between two competing products, something closer to how a calendar app stays matched across two phones. Anthropic's own documentation never describes this feature as a sync, and for good reason. Nothing about the mechanism keeps running after you close the import screen.
What Import Memory Actually Does
Anthropic's own support page lays out the mechanism in a short sequence of manual steps, and none of them involve Claude reaching into ChatGPT's servers on its own. First, you export from your current AI assistant: Anthropic provides a prompt you paste into ChatGPT, Gemini, or whichever tool you're leaving, asking it to list everything it has stored about you in a single block of text. Second, you open the import flow inside Claude, under Settings, then Memory, and select "Start import." Third, you paste the text you copied and click "Add to memory." Claude then reads that pasted block once and files pieces of it away as individual memory entries you can review afterward.
- Export: run a prompt inside your old assistant, usually ChatGPT or Gemini, asking it to write out everything it remembers about you in one block of text.
- Import: inside Claude, go to Settings, then Memory, and choose "Start import," which opens a text box for that exported block.
- Submit: paste the text and click "Add to memory." Claude reads it once and stores pieces of it as separate memory entries.
- Review: open the memory panel afterward to see exactly what Claude kept, and edit or delete any entry by hand.
Anthropic labels the whole feature experimental. Its support page states plainly that memory imports are "experimental and still in active development," and that Claude's memory "may not retain imported personal details unrelated to work," since Claude's memory system is built to focus on work topics. Neither caveat describes a syncing relationship. Both describe a one-time ingestion step that the receiving model is still learning to handle well.
The same support page also notes that users on Claude's older memory experience reach the same tool from a different spot, under Settings, then Capabilities, or from an "Import memory to Claude" card on the home screen. Different path, same underlying step: paste a block of text once, click a button once, and let Claude sort it into entries you can check afterward. Anthropic built two doorways into the feature, but both doorways lead to the identical one-time action.
Why "Sync" Is the Wrong Word
A sync, in the ordinary sense of the word, is a standing relationship between two systems that keeps both of them current without a person repeating the same action. Email syncs across a phone and a laptop because both devices stay connected to one server, checking for changes on their own. Nothing like that exists between Claude and ChatGPT. Once you paste a memory export into Claude and click "Add to memory," the two products go back to knowing nothing about each other. Claude is not watching ChatGPT's memory for changes, because Claude has no access to ChatGPT's memory at all, only to whatever text a person manually copied out of it on one occasion. If you want Claude to see something new you told ChatGPT last week, you have to repeat the entire export-and-paste process yourself.
What "Going Stale" Actually Means Here
A memory snapshot is accurate at the moment you create it and starts drifting from the truth the moment anything changes afterward. Say you told ChatGPT in January that you were using a specific tool at work, then switched teams in April and started using something else entirely. If you exported your ChatGPT memory back in February and pasted it into Claude, Claude is still working from the January version in October, unless you go back and run the export again yourself. ChatGPT's own copy would have kept updating as you kept using it. The copy sitting inside Claude would not, because nothing tells it to. The same thing happens with smaller facts nobody thinks to double-check: a phone number you no longer use, a diet restriction you dropped months ago, a client you stopped working with. None of those get flagged as outdated. They just sit there, treated as current until a person notices Claude repeating something that stopped being true.
The gap between a fresh memory and an exported one shows up quickly in practice. One writer who reviewed her own stored ChatGPT memory before migrating it to Claude found that most of it no longer matched reality. She estimated she would keep only 20 to 30 percent of what was actually stored, because, in her words, most of it was outdated and some of it was simply irrelevant by the time she looked at it again. That staleness existed inside ChatGPT's own memory before any import ever happened. Pasting it into Claude does not refresh it. It just moves the same outdated snapshot into a second place.
An imported memory is a photograph, not a window. It shows you what was true when you took the picture. It will never show you what changed after that, no matter how long you leave it sitting in the new assistant's memory panel.
What Real Continuous Sync Would Actually Require
A real sync would need a live, two-way channel both companies agree to keep open, an automatic trigger whenever either assistant's memory changes, and a shared format for resolving the moment both products hold conflicting versions of the same fact. It would also need each company to grant the other standing, ongoing access to user data inside its own product, something OpenAI, Anthropic, and Google have no business reason to build for a rival's chatbot: a memory import feature exists to pull a customer away from a competitor, not to make that competitor's product more convenient to keep using alongside it. Treating a one-way, one-time doorway as a sync between equals misreads both the mechanism and the incentive behind why it was built.
One-Time Import vs Continuous External Memory
| What to check | A one-time import | Continuous external memory |
|---|---|---|
| How it starts | A person manually runs an export prompt in one assistant and pastes the result into another | You add context directly, in whichever tool you're using, without a separate export step |
| What keeps it current | Nothing. The pasted text is frozen the moment you click "Add to memory" | The layer itself, since there's no separate snapshot sitting apart from the source |
| Direction | One way, usually only into the newer product trying to win you over | Available to any connected tool, not built to pull you off one product |
| What happens when the original changes | The imported copy has no idea. You have to export and paste again by hand to catch up | The next time you ask, it reflects what's actually there |
| Who has to do the repeat work | You, every time you want the copy refreshed | No one. There's no copy sitting separately to refresh |
Before relying on an "import memory" feature for anything that changes often, like your current job, your active projects, or your health details, ask yourself when you last refreshed it. If the answer is longer ago than you'd trust from memory alone, treat the imported copy as a starting point, not a current record.
What to Do With an Import Instead of Expecting a Sync
The broader question of what AI memory import features actually move, and why the copy they leave behind decays, is its own topic; this piece stays narrowly on the specific "syncs with ChatGPT" claim and what Claude's current import screen does and does not do. Used correctly, an import feature like Claude's is a head start, not a permanent bridge. It saves you from retyping years of context into a blank chat the first time you try a new assistant. It does not save you from repeating that export whenever the original has moved on. Anyone planning to depend on it for more than a first setup should expect to run the export prompt again every few months, check what actually got kept in the memory panel, and delete anything that no longer matches reality rather than let an assistant act on a stale fact with full confidence. Treat the first import as a one-time favor the feature does for you, not a subscription to anything ongoing, and the whole feature stops being confusing.
This is the specific gap a one-time copy cannot close, and it is the problem MemX is built around. MemX (memx.app) is an external memory layer that recalls context across ChatGPT, Claude, Gemini, and a person's own documents, and it stays current because there is no export step to forget to repeat. You add to it as you go, in whichever tool you happen to be using that day, instead of taking a single snapshot and carrying it from one assistant to the next. MemX is private by architecture, with per-user isolation and encryption at rest for whatever it holds. It does not sync Claude's internal memory with ChatGPT's, and no product on the market does that today. What it does is give a person one place their own context actually lives, so a fact does not go stale the moment they close the tab where they first said it.
01Does Claude's memory actually sync with ChatGPT?
No. Claude's import feature is a one-time, manual copy. You ask ChatGPT to summarize what it remembers about you, paste that text into Claude's import screen, and Claude stores pieces of it. Nothing updates automatically afterward if your ChatGPT memory later changes.
02How do you import ChatGPT memory into Claude?
In ChatGPT, run the prompt Anthropic provides to list your stored memories, and copy the result. In Claude, go to Settings, then Memory, select "Start import," paste the copied text, and click "Add to memory." Claude extracts and stores pieces of it for you to review.
03Does imported memory update automatically after you import it?
No. Once pasted, the memory sits in Claude as a fixed snapshot. If you later tell ChatGPT something new, Claude never sees it unless you manually repeat the export and import steps yourself. Nothing refreshes on its own.
04What does it mean when AI memory "goes stale"?
It means saved information stopped matching reality after something changed, like a job, a project, or a preference, but the assistant keeps treating the old version as current. An imported snapshot is especially prone to this, since nothing updates it after the one-time paste.
05Is there an AI tool that syncs memory across ChatGPT, Claude, and Gemini automatically?
Not between the chat products themselves. Each treats its own memory as separate, and none maintain a live connection to a competitor's memory. An external memory layer you update directly, instead of exporting and re-pasting, is the closest thing to that today.
