You spent a year teaching ChatGPT how you write, what your projects are called, which framework you prefer, and that you hate being asked clarifying questions. Now you want to try Claude or Gemini for a week. Obvious question: can you bring all that memory with you?
Sort of. One direction at a time. And not really. In March 2026 all three big assistants added memory portability within weeks of each other. The marketing made it sound like you could pick up your AI brain and set it down somewhere new. What you actually get is a photocopy of one page of notes, carried across the street by hand. This post walks through what each platform lets you move today, how to do it, and the catch the headlines skip.
What "moving your memory" actually means
One distinction explains everything below. People use "AI memory" for two very different things:
- Chat history: the raw transcripts of every conversation you have had. Bulky, exportable, and usually delivered as a ZIP of JSON files.
- Saved memory: the short profile the assistant has distilled about you. Your name, job, tone, recurring projects, stated preferences. Small, valuable, and the part that makes the assistant feel like it knows you.
The 2026 tools mostly move a text summary of the second kind, or a bulk dump of the first. Neither moves the assistant's continuously updated model of you. Hold that split in your head. Every catch below comes from it.
Gemini: two import tools, one big asterisk
Google shipped the most complete pair of tools on March 26, 2026. Two separate paths, two different jobs.
1. The copy-paste preference prompt
Gemini hands you a block of text. You paste that prompt into ChatGPT or Claude, the other assistant writes out a summary of what it remembers about you, and you copy that summary back into Gemini. That is the memory transfer. A one-time snapshot in plain language. Not a sync.
2. The chat-history ZIP import
For the bulky kind, Gemini ingests the standard ZIP export you download from ChatGPT or Claude. Files up to 5 GB are accepted. You can upload up to five ZIPs per day, and re-uploading adds new chats while overwriting your previous import. This hands Gemini your transcripts to learn from, which is not the same as importing the other assistant's distilled memory.
Regional catch: both Gemini import tools are unavailable in the EEA, UK and Switzerland. If you are in those regions, this entire path is closed to you, no matter which assistant you are leaving.
Claude: the paste-into-ChatGPT extraction prompt
Anthropic got there first, rolling memory import out around March 2, 2026, and bringing memory to the free tier at the same time. The mechanism mirrors Gemini's prompt method. You go to Claude's import page, copy a prompt Anthropic wrote, paste it into ChatGPT (or Gemini, or Grok), let that assistant produce a formatted list of your memories, then paste the result into Claude's memory settings.
Anthropic is honest about what this is. It saves you from rebuilding your preferences from zero after switching. A migration aid, not a bridge. Once the text lands in Claude, Claude owns its own copy and starts editing it based on your conversations with Claude. The original keeps living its own life inside ChatGPT.
ChatGPT: export exists, but mind the gap
OpenAI added a memory export in March 2026, producing a JSON file of stored facts and preferences. Two things matter. There is no API for exporting memories, so you cannot automate it. And historically the saved memories lived in a separate store that did not come out in the standard Settings > Data Controls > Export data ZIP, which only held your chat transcripts. So when you ask ChatGPT to export, be clear about which memory type you are getting. They have lived in different places.
OpenAI moved the opposite way on visibility. A May 2026 update added Memory Sources, which shows what ChatGPT referenced for a given answer: past chats, saved memories, custom instructions, your file library, and connected accounts like Gmail. Good for trust. It makes none of it portable.
The one catch that ties it all together
Read the three sections above and the same shape appears every time. You move a one-time text transfer of summaries, never full context. The moment the import finishes, the two sides drift apart. Tell Claude about your new job and ChatGPT will never hear about it. Update your preferences in Gemini and Claude keeps the stale version forever.
There is no native any-to-any path either. ChatGPT memories do not natively transfer into Claude. Claude memories do not natively transfer into Gemini. The prompt-based tools work by asking the source assistant to write a summary in plain English, which is exactly why the result is lossy and frozen. You are not connecting two memory systems. You are asking one to dictate notes that the other transcribes once.
| Platform | What you can export | How you import it elsewhere | The catch |
|---|---|---|---|
| ChatGPT | JSON of saved facts/preferences (Mar 2026); chat history as a ZIP | Paste it as a summary into Claude or Gemini, or feed the ZIP to Gemini | No export API; saved memories historically sat outside the standard data export |
| Claude | Full memory text, exportable anytime; chat history ZIP | Paste into Gemini's import box, or upload the ZIP to Gemini | Import is a one-time paste; afterward Claude's copy drifts from the original |
| Gemini | Relies on a copy-paste prompt plus ZIP ingestion from rivals | Two tools: preference prompt and chat-history ZIP import | Both tools blocked in EEA, UK and Switzerland; ingesting transcripts is not the same as importing distilled memory |
Why every assistant suddenly cares about remembering you
The timing is not a coincidence. Each assistant remembers in its own style. ChatGPT runs a personal-assistant memory built from saved memories plus chat history. Claude leans toward project and organizational memory, and brought that memory to its free tier on March 2, 2026 after rolling it out to paid tiers in late 2025. Gemini frames its version as Personal Intelligence across the Google ecosystem, with automatic memory and privacy controls in place since 2025.
The models also keep leaping. GPT-5.5, Gemini 3.5 and Claude Opus 4.8 all landed in this window. Then at WWDC on June 8, 2026, Apple unveiled a rebuilt Siri AI running on a custom version of Google's Gemini. Read that again. The model behind your phone can now swap overnight. When that happens, the thing that keeps you loyal is no longer the model. It is whether it still remembers you. Which is exactly why each company wants the import door to open inward and the export door to feel a little stiff.
The contrarian read: portability is mostly marketing
As long as your memory lives inside an assistant, that assistant owns the canonical copy and every other tool gets a frozen photocopy. Import buttons do not change this. They make one specific switch hurt less, and that switch conveniently points toward the company that built the button. The lock-in was never your chat logs. It is the version of you that you cannot extract or keep in sync. That is the lock-in.
The only durable fix is structural. Keep your memory outside any single assistant, in a place you control, and feed it to whichever model you happen to be using this month. If your memory is the canonical copy and the assistants are interchangeable readers of it, the import-button race stops mattering to you.
Where MemX fits
This is the gap MemX is built for. Instead of letting your context pile up inside one assistant, you keep it in one place you own. You capture your own documents, photos, voice notes and chats, then ask questions in plain English and get answers with citations back to the source, plus natural-language reminders. Because the memory lives with you, not inside a vendor, switching assistants stops meaning starting over.
On privacy, MemX is private by architecture: customer-managed encryption keys (CMEK), per-user isolation, encryption at rest with AES-256-GCM, and on-device options. To be plain about it, that is not the same as end-to-end encrypted or zero-knowledge, and we do not claim to be either. The point is ownership and isolation. Your memory is yours, kept separate, and never pooled into someone else's model.
Before you migrate anything, export from your current assistant first and save the file. Import tools overwrite. Re-uploading a ZIP to Gemini, for example, replaces your previous import. A local copy is the only version next month's update cannot silently overwrite.
So, can you move your AI memory between apps?
You can move a snapshot. You cannot move the living thing, and you cannot keep two assistants in sync. Use the import tools for what they are good at, which is dodging a cold start when you switch. Just do not mistake a one-time paste for true portability. If you want memory that genuinely follows you across ChatGPT, Claude, Gemini and whatever powers your phone next year, the answer is not a better import button. It is keeping the canonical copy outside all of them.
01Can I transfer ChatGPT memory to Claude?
Indirectly. Claude gives you a prompt to paste into ChatGPT, which writes a summary of what it remembers. You paste that into Claude. There is no native, automatic transfer, and the copy stops updating once imported.
02Does Gemini's import work everywhere?
No. Gemini's preference-prompt and chat-history import tools launched March 26, 2026 but are unavailable in the EEA, UK and Switzerland. Users in those regions cannot use either path right now.
03Does exporting ChatGPT data include my saved memories?
Not reliably. The standard data export historically held chat transcripts, while saved memories sat in a separate store. OpenAI added a memory export in March 2026, but there is no API and the two have lived in different places.
04Will my imported memory stay in sync between apps?
No. Every import is a one-time text snapshot. The moment it lands, the two assistants evolve independently. Update one and the other keeps the old version forever. No live sync exists between any of them.
05How do I keep my AI memory portable for good?
Stop storing it inside one assistant. Keep your context in a place you own, like MemX, and feed it to whichever model you use. Then switching assistants does not mean rebuilding your memory from scratch.
Written by Arpit Tripathi, who works on AI memory at MemX.
