Every claim on this page comes from the vendor's own help documentation, read on 4 October 2026 and linked in the section where it appears. Memory features change month to month, so treat that date as the honest expiry.
What “memory” means in each product
ChatGPT: one summary that rewrites itself
ChatGPT's improved memory draws on your past chats, saved memories, custom instructions, files in your Library and connected apps such as Gmail (which of these apply depends on your plan and region), and keeps a continually updated summary of that context. The memory summary you see is a high-level view, not a full list of everything Memory can reference; when personal context shapes an answer, a Sources list may appear under it. You read it in Settings › Personalization › Memory, edit it by typing a change into the text box or highlighting a passage, and remove it with “Delete and turn off memory”, which leaves your past chats in place. The older saved memories notepad — the list of discrete facts ChatGPT used to keep — is now labelled legacy; you can switch back to it from the same screen. Temporary Chats never create or update memory; before you start one you choose whether it may use your existing memory, or pick Unpersonalized to keep memory out entirely, and Custom Instructions remain a separate control. Source: OpenAI: Memory in ChatGPT. Full walkthrough: how ChatGPT memory works
Claude: a list of topics you can open
Claude saves memory as individual topics while you chat — a moved deadline, a new project — rather than summarising conversations after they end, and you can tell it “remember this” to save something directly. Memory is on by default on the Free, Pro and Max plans across web, desktop and mobile, while on Team and Enterprise an owner decides whether it is available, and each member then turns it on. Each project has its own memory space. Settings › Memory lists every topic; you edit or delete them one at a time, pause memory (keep it, stop using it) or reset it (delete everything). Sensitive topics are left out unless you opt in, incognito chats are never remembered, and searching past chats — a separate feature — is limited to paid plans. Claude also documents an import and export path and labels it experimental. Sources: Claude Help Center on memory, import and export your memory. Full walkthrough: does Claude have memory?
Gemini: your past chats, read back
Gemini's memory of past chats sits under Settings & help › Personal Intelligence. It needs a personal Google Account rather than a work, school or supervised one, an adult account holder, and the Keep Activity setting on; it works in the Gemini app, on the web, in Gemini in Chrome and on a smartwatch, but not inside Gems or Live conversations. There is no list of memories to edit: you correct Gemini in the chat itself, and to make it forget something you delete the chats that contain it from Gemini Apps Activity, with a short delay before the deletion takes effect. Saved info, Gemini's standing instructions, is a separate feature. Source: Google Gemini Apps Help on memory. Full walkthrough: does Gemini have memory like ChatGPT?
Side by side
Where a vendor's page does not address a row, the cell says so rather than guessing.
| Question | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Where the memory comes from | Past chats, saved memories, custom instructions, Library files and connected apps (varies by plan and region), kept current in a continually updated memory summary. | Your chats, saved as individual topics while you talk; “remember this” saves something directly. | Your past chats, plus connected Google apps when Personal Intelligence is on. |
| Who has it | Anyone with Memory on in Settings › Personalization › Memory; OpenAI says features and controls vary by plan, region, platform and workspace settings. | On by default: Free, Pro, Max. Team and Enterprise: an owner makes it available, then each member turns it on. | Personal Google Accounts only, adult account holders, Keep Activity on. Not in Gems or Live. |
| See and edit it | Read the summary; change it by typing into the text box or highlighting a passage. Legacy saved memories are a discrete list. | Settings › Memory lists every topic; edit or delete each one, or tell Claude in chat what to change. | No memory list. Correct Gemini in the conversation itself. |
| Delete or turn it off | “Delete and turn off memory” from the three-dot menu; past chats are not deleted with it. | Delete a topic; Pause memory (keeps it, stops using it); Reset memory (deletes everything, irreversible). | Turn Memory off under Personal Intelligence; delete the chats that hold the information from Gemini Apps Activity. |
| A chat it will not remember | Temporary Chat: creates no memory; choose Unpersonalized before starting to also stop it using existing memory. | Incognito chat — not saved to memory or to chat history. | Temporary chat — described in Google's privacy hub as the way to discuss what you don't want Gemini to remember. |
| Take it with you | The Memory FAQ describes no export or import of memory. | Yes — import from other assistants and export for backup, documented as experimental. | Import only: memories (paste) and full chat history (.zip) from other assistants; no memory export documented. Not available in the EEA, Switzerland or the UK. |
| Readable by another AI tool | No. | No, except by exporting and pasting. | No. |
What survives a deleted chat
ChatGPT. Legacy saved memories live in a notepad kept separately from chat history, so deleting a chat does not delete the memories it produced. The memory summary is rebuilt from what remains; deleting the summary does not delete past chats, and if you turn memory back on later ChatGPT may build a new summary from the chats still in your history.
Claude. Deleting a conversation does not remove the memory topics generated from it — you delete topics individually, or reset everything. Memory is included in your data export.
Gemini. The reverse arrangement: the memory is the past chats. Delete the chats that contain the information and Gemini stops using it after a short delay; information that arrived through a connected Google app also needs that app disconnected, or Gemini can find it again.
Does any of them fill up?
Users of ChatGPT's saved memories have reported a “Memory is full” notice on OpenAI's own forum for years. OpenAI's current FAQ describes no capacity for the memory summary and treats the notepad as legacy, so if you still see the notice you are on saved memories: prune the list, or switch to the improved memory from Settings › Personalization › Memory. Neither Anthropic's nor Google's page describes a memory capacity at all.
Portability: the row none of them wins
Line the three up on the question that decides whether memory is an asset or a liability — can it leave? — and none of them lets it leave cleanly. ChatGPT documents no way to export memory. Gemini documents a way in — an “Import memory to Gemini” flow for pasted memories and uploaded chat-history .zip files (Google's import help page) — but no way out. Claude documents an import and export flow and is candid that it is experimental: you ask Claude to write out its memories verbatim, and you paste memories from another assistant into an import box. That candour is worth something. It is still copy-and-paste, and it moves the memory into another silo rather than out of silos altogether.
The alternative is to keep the knowledge that matters outside all three, in a store each of them can read. That is what an AI memory layer is: a knowledge base with an MCP server in front of it, connected once per assistant. BrainTube is ours, so read this paragraph as labelled. It holds what you save — YouTube, podcasts, articles, PDFs — and makes it queryable from Claude, Cursor and other MCP clients (ChatGPT support in testing); it will also hold the Claude and ChatGPT conversations you choose to save into it, it does not record your chats in the background, and it is not a replacement for the profile-style memory described above. For which clients share one memory today, see one memory across ChatGPT, Claude and Cursor; for the third-party tools, the AI memory tools roundup; for setup, the MCP memory guide; for leaving ChatGPT, ChatGPT memory alternatives.
So which AI has the best memory?
- Most automatic: a draw. All three now learn from your chats without being asked — ChatGPT into a summary, Claude into topics, Gemini by reading past chats when Memory is on.
- Most visible and editable: Claude, narrowly. A per-topic list you can open, edit and delete one at a time is the default there; ChatGPT's default is a single editable summary, with a discrete list only if you revert to legacy saved memories; Gemini offers no list at all.
- Easiest to keep a conversation out of it: a draw. ChatGPT's Temporary Chat (set to Unpersonalized), Claude's incognito chat and Gemini's temporary chat all do the job.
- Most portable: Claude, with the asterisk that the export is prompt-based and experimental. Gemini can import memories and chat history but documents no export; ChatGPT documents neither.
- If you use more than one assistant: none of them. Each memory is readable only inside the product that made it; a shared store outside all three is the only arrangement where the effort compounds.
The one-line answer: Claude currently documents the most control and the only exit, ChatGPT's summary is the most hands-off, and Gemini's is the most tied to a Google account. Which is “best” depends on whether you value control, convenience or portability — and on how many assistants you actually use.
How these claims were checked
Every vendor claim on this page was read from the vendor's own help documentation on 4 October 2026 and is linked where it appears. Where a page does not address a question, the table says so instead of estimating. No prices, plan quotas or usage figures appear here on purpose: they change faster than a page like this can be re-checked.