How to Transfer Your AI Memory Between ChatGPT, Claude, Gemini, and Grok (2026 Guide)

1. Moving Without Losing Your Luggage

For two years, you've been training ChatGPT. It has finally learned that you're (still) not a vegetarian, that you're writing a dissertation on the Vikings, and that you can't stand it when a reply starts with "Of course!" or, worse, "Great question." Claude, Gemini, and the new Chinese models sound intriguing, but the fear of having to walk that road all over again is enough to kill your curiosity and any desire to try something new.

Good news: those fears are unfounded. Your memory is luggage, not furniture. Packing your suitcase takes about ten minutes, a plain text file, and not a single line of code, we promise.

2. What Is "Model Memory," Anyway?

It's worth sorting out the terminology around model memory right away.

Here's the key thing: the neural network itself doesn't remember you at all. Between any two of your requests, it's a perfect blank slate, and it has no file on you stored inside it (unless, of course, there's a Wikipedia article about you that it was trained on). Yet a memory of you definitely exists and is stored somewhere associated with the network, which is precisely what keeps you from switching to other services. How do we resolve that paradox? It’s surprisingly simple.

Your dossier doesn't live in the model itself, but in its chat assistant, a layer on top of the LLM that provides memory, adjustable settings, and various other bells and whistles. You can't work directly with the bare model. When you think you're writing to it, you're actually addressing the assistant, which translates your questions into language the software understands. Assistants can be native or independent, connected in the latter case with an API key, and advanced wrapper apps let you carry over your entire history. Even if you download a neural network onto your own computer, you'll still need to set up an assistant separately and import your memories from previous models into it.

Here’s the rough terminology. “The model,” ”the base model,” or “the foundation model” refers to the LLM itself, the software that stores no memories of you whatsoever. “The AI assistant,” “the chat assistant,” “the app,” “the application layer,” or “the product layer” means the wrapper that remembers you. Experts refer to this as scaffolding or harness when they want to emphasize that it’s a technical layer.

The memory feature itself is called something similar across all these products:

• ChatGPT — Memory, which covers two mechanisms: Reference saved memories and Reference chat history.
• Claude — Memory.
• Gemini — Personal Intelligence, with Memory tucked inside it. Google is renaming these features as we speak, so don't be surprised if the menu labels shift right in front of you.
• Grok — Memory.

Under the hood, the feature is refreshingly unglamorous: a plain database of text notes about you that the harness slips in front of the LLM at the start of every new conversation. The model is a brilliant but amnesiac consultant; the harness is the secretary who places a folder with your dossier on their desk before every meeting.

AI uses two types of "memory," and it's important not to mix them up.

Context memory (also called in-chat memory) is the amount of text the model can hold in view at once. While the conversation is short, the LLM remembers all of its contents. Once the chat grows long enough, things get ugly: older messages get pushed out and the model starts mixing things up, forgets agreements you made earlier, and generally starts acting like someone four hours into a meeting that should've ended long ago. A long chat is always a stress test for context memory. (More on this: LLM Context Window: How 10 Million Tokens Fool You.)

Persistent memory about you (also called permanent memory) is that folder on the secretary's desk. It doesn't vanish when you close the chat — it gets mixed into every new conversation. It typically has two layers: saved memories, an explicit list of facts you can open in settings and read, and a summary (or profile), which the harness compiles from your entire chat history and usually doesn't tell you a thing about. This is the folder — in its entirety, every layer of it — that we're about to move.

3. The Memory File: Packing Your Bags

We're going to move your memory the lowest-tech way possible: with a plain text file. Let's call it the memory file. It doesn’t have any fancy formats, just a simple document (.md beats .txt), containing a structured list of everything the model knows about you — who you are, what you do, what you like, what you can't stand, what you're working on.

You get it with one universal prompt that works almost everywhere:

Write out, verbatim, everything you remember about me: all saved memories and everything you know about me from our chat history. Don't shorten, summarize, or omit anything. Put it all in a single text block so I can copy it.

The key word here is "verbatim." Leave it out and the model will start waxing lyrical ("the user is a pleasant person with wide-ranging interests"). What we need right now isn't a memoir, it's an inventory.

Save the resulting text to a file on your computer. While you're at it, read it over. First, you'll learn a lot of new things about yourself. Second, this is the only copy of your digital portrait that you personally own. Admit it: that feels good.

Let's figure out how to pull this luggage out of each specific model.

4. Export Guide: How to Get Your Memory Out of ChatGPT, Claude, Gemini, and Grok

The universal prompt from the previous section will let you get your memory from all of the models on our list. Below you’ll find a quick rundown of each LLM’s quirks in this regard and a precise step-by-step guide to the process of memory retrieval.

ChatGPT

The main “gotcha” here is that the official-looking Export data button will send you an archive with your entire chat history, but no memory of you. Work with the prompt instead, as it pulls out both the saved memories and whatever the model has figured out about you from your chat history.

Exporting from ChatGPT:

1. Start a new chat (in a browser, ideally, since copying a long inventory from your phone is its own kind of misery).
2. Paste the universal prompt from Section 3. Send it.
3. Copy the entire response.
4. Paste it into a text file and save it.
5. Double-check by looking in Settings → Personalization → Memory to verify everything made it into the file. Manually add any missing elements.

Claude

Here everything is suspiciously transparent: the memory is fully visible in the settings and Anthropic itself officially recommends the prompt.

Exporting from Claude:

1. Open Settings → Capabilities → Memory → "View and edit your memory."
2. Select all the text and copy it.
3. Paste it into a text file and save it.

You can also use the same universal prompt in a new chat. The result is just as good.

Gemini

Gemini’s secretary keeps two folders but only shows you one: while the explicit entries are visible in the settings, Google doesn't show the hidden summary of your chats anywhere. Google Takeout won't help — it'll export your chats, but not the memory. Here's how to pull both folders out:

Exporting from Gemini:

1. Open Settings & help → Personal Intelligence. You'll need a personal Google account, as the feature isn't available on work or school accounts. Also, Google is rolling Personal Intelligence out gradually: if you can't find the option, don't panic, just check back in a week.
2. Copy all the entries from your instructions for Gemini into a text file.
3. Start a new chat, paste the universal prompt, and send it.
4. Add the response to the same file.
5. Read it over: Gemini has a habit of returning incomplete lists. If you know something that didn't make it in, add it yourself. You're allowed.

Grok

Grok’s memory is visible as a list in the settings, but the official export (a ZIP with your chats) doesn't include it. Here's the juicy part: the "Turn off memory" button doesn't delete your data. It just pauses recording, and the entire dossier you've accumulated stays sitting in the settings until you delete it manually.

Exporting from Grok:

1. Profile icon → Settings → Data Controls → Memory.
2. Copy the list of memories into a text file.
3. As with Gemini, make a new chat and paste the universal prompt, then add whatever wasn’t in the list to the file.

Now you've got a memory file in hand — maybe more than one, if you've been chatting with several bots at once. All that's left is to get the luggage through customs.

5. Import Guide: How to Import Your Memory into Claude, ChatGPT, Gemini, and Grok

You've got the memory file. Now let's teach the new model to read it. The general principle is the same everywhere: paste the text and ask the LLM to remember what you wrote.

Claude

Claude was the first to set up an official import procedure. Anthropic built it right into settings, complete with a ready-made prompt for extracting memory from ChatGPT, Gemini, and Grok (yes, the same trick we covered in Section 3 — nice to know the industry agrees with us). The feature is marked experimental, which makes the last step all the more important.

Importing into Claude:

1. Open Settings → Capabilities → Memory → "Start import."
2. Paste the contents of your memory file into the import field.
3. Click "Add to memory."
4. Processing the file takes up to a day. A day later, open "View and edit your memory" and check which facts the model extracted. If something got lost, add it by hand right there in the menu.

ChatGPT

No official import option exists, but you can carry your “luggage” through two “doors”: a regular chat (for uploading facts) and the Custom Instructions menu (a limit of around 1,500 characters per field).

Importing into ChatGPT:

1. New chat → paste the memory file → add: "Save this to memory as facts about me. Don't shorten anything."
2. Check: Settings → Personalization → Memory. ChatGPT likes to creatively recompress your data, so feed it whatever's missing again, chopped into smaller pieces.
3. Duplicate the most important stuff (response style, profession, restrictions) in Settings → Personalization → Custom Instructions. It's more reliable than the memory itself.

How to Import Memory into Gemini

Google answered Anthropic symmetrically and then some. It also offers two elements, but while the first serves a similar memory file to that of ChatGPT, the second compresses your entire chat history into a ZIP archive. What's more, Google itself hands out a prompt for extracting memory from ChatGPT and Claude: our method from Section 3 has now been officially crowned an industry standard twice over.

An important note on privacy: Google may use anything you import into Gemini to improve its services, including training models. Give your dossier a careful re-read before uploading it.

Importing into Gemini:

1. Open Settings & Help → Import Memory to Gemini.
2. Paste the contents of your memory file into the field → "Add Memory." Gemini will start a new chat and sort the facts onto its shelves.
3. Want to bring over the actual conversations too? In the same place, under "Import Chats," click Add and upload the ZIP from your old assistant's official export (up to 5 GB, up to five archives a day). Imported chats will show up in the sidebar with a special marker.
4. Check back a day later: processing can take up to 24 hours.

Grok

xAI's official import is still in testing, so for now you'll have to do everything by hand.

Importing into Grok:

1. New chat → paste the file → "Remember this about me."
2. Check the list: Settings → Data Controls → Memory.
3. Duplicate the key points in Custom Instructions. xAI has changed this field's limit more than once, so it's best not to cram your entire life story in there. Just put the most important stuff first. Remember that information about Grok goes stale at an incredible rate, even by the standards of the AI era.

Qwen, DeepSeek, Kimi: Memory in the Chinese Models

With the residents of Chinatown, things are a lot more straightforward.

Qwen. This LLM’s memory works like Grok's. For export, copy the list from settings and drop the prompt into a new chat; for import, paste the file into a new chat and tell it "Remember this."

DeepSeek and Kimi. Neither of these models retain persistent memory of you. Kimi has a giant context window, but that runs on memory within a specific conversation and doesn’t give you anything to export. Importing requires you to paste your text file at the start of every new chat. This isn’t very convenient, but it is cheap — prices in Chinatown are an order of magnitude lower than they are for western LLMs.

Xiaomi MiMo. There's almost no point asking about memory with MiMo: it lives mostly in the API, and it has no chat assistant. The only harness with its own memory is the console-based MiMo Code, but that's a tool for programmers.

6. The Fine Print

You're better off knowing three things in advance.

First, geography plays a major role. Grok's memory doesn't work in the EU and the UK, and Gemini's import isn't available in the EU, the UK, and Switzerland.

Second, incognito and temporary chats (all of the big three have them) don't write anything to memory. This is handy when you want to ask something dumb without consequences for your dossier.

Third, any import is a retelling, not a copy. The model may compress, paraphrase, or drop a chunk of writing (and images and files from old chats won't carry over at all — only text), so the "check what it actually remembered" step isn't paranoia, just healthy skepticism.

One final piece of advice: don't throw away your memory file after the move. Update it every couple of months with the same prompt so that your next switch will start straight from the place where you left off with the other model. Plus, your “luggage” becomes a plain text file that you can always edit, trim, or add to by hand — if you can be bothered.

Memory Export and Import at a Glance

7. Finale: The Luggage Matters More Than the Hotel

Models will keep changing faster than you can get attached to them. Today's favorite might fall hopelessly behind tomorrow, and that's fine: as we agreed at the very start, AI apps are just hotels. The real asset — your packed luggage — stays with you no matter what. That modest little memory file holds far more of you than any cloud ever could. Keep it close, keep adding to it, and any future move will take you all of a few minutes.

Don’t attach too much importance to the new model’s knowledge of you. You're not arriving as a newborn, but as a fully-formed adult, with a suitcase full of self-knowledge.