Blog
ChatGPT vs Claude vs Gemini memory: what each remembers
All three big assistants now remember you between chats. They differ in what they keep, how much you can see and edit, and whether you can take it with you. Here’s a side-by-side from each company’s own help pages, and the one thing none of them remembers.
From each company’s help pages, checked October 21, 2026. Plan details change often; follow the links for the current terms.
Side by side
| ChatGPT | Claude | Gemini | |
|---|---|---|---|
| Saves facts about you | Saved memories | Memory | Saved info (Personal context) |
| Uses past chats | Reference chat history | Chat search and memory | Past chats (rolling out on mobile) |
| See and edit what it knows | Settings › Personalization | Settings › Memory | Settings & help › Personal context |
| Separate memory per project | Projects can keep their own | Projects keep their own context | Not described |
| Chats that don’t count | Temporary Chat | Incognito chats | Temporary Chats |
| Import from another assistant | Not described | Yes, experimental | Not described |
| Knows work outside the chat | No | No | No |
ChatGPT
ChatGPT has two kinds of memory, both under Settings › Personalization. Saved memories are facts it keeps about you, which you can view and delete, and which can fill up (see ChatGPT “memory full”). Reference chat history lets it draw on past conversations without saving them as facts. Projects can be set to keep their memory to themselves, and Temporary Chats don’t use or add to memory. OpenAI’s memory FAQ lists what each plan gets.
Claude
Claude can search your past chats and keeps a memory of what it learns about you, which you can view and edit in Settings › Memory. Projects hold their own instructions and files, and incognito chats aren’t remembered. Claude is the only one of the three with an import for memories from another assistant, which Anthropic calls experimental (see switching from ChatGPT to Claude). Which plans have what is in Anthropic’s memory help.
Gemini
Gemini calls it Personal context, under Settings & help. It holds saved info (things you asked it to remember, like “always give me vegetarian recipes”) and, where available, the option to reference past chats, which Google says is still rolling out in the mobile app. It requires signing in, isn’t available to users under 18, and doesn’t apply in Gemini Live. You can also manage saved info at gemini.google.com/saved-info. See Google’s help page.
What none of them sees
All three remember what happened in the chat. None of them knows what you did outside it: the call this morning, the document you edited, the decision in Slack, the code your agent wrote. So the most useful context at work, what happened recently, is exactly what you end up pasting in, in all three. More in why AI memory goes stale.
Memory that goes with you
If you use more than one assistant, or might switch, keep work context somewhere all of them can read. An MCP memory server does this. Cachejay records your workday on your Mac (screen text, calls, coding-agent sessions) and files it by project, and each assistant connects differently:
- Claude: on the web, desktop and phone. Setup.
- ChatGPT: on the web, on Business, Enterprise and Edu plans; read-only on Pro; not on Plus or Free. Setup.
- Gemini: the Gemini app connects through the cloud, set up in your browser, and Antigravity connects on the Mac. Setup.
The assistants keep remembering your preferences; the shared memory covers what happened. See also give Claude memory of your work and the AI memory privacy checklist.