Alcoven

Does ChatGPT remember across chats?

Blog · Published 14 October 2026

Yes, for you. With memory on, ChatGPT carries what it learns into new chats in two ways: saved memories (details you asked it to keep, or ones it kept on its own) and your chat history, which it looks through when that would help a reply. Both are tied to your account: a colleague’s ChatGPT doesn’t see them, and neither does Claude or any other AI. Projects can keep a memory of their own, and in a shared project ChatGPT draws on every member’s chats in it.

Here is how each part works, how to manage it, and where it stops, as OpenAI’s help pages describe it today.

How does ChatGPT remember other chats?

Saved memories are what ChatGPT keeps on purpose. Chat history is what it looks up. OpenAI describes and controls them separately:

Saved memories Chat history
What it is Details you asked ChatGPT to remember, or that it saved as useful Relevant context from your past conversations
How it’s used Carried into new chats Looked up when it’s likely to improve a reply
Limit Not stated on OpenAI’s current memory page (see “memory full” below) No separate storage limit
When you turn it off Turning off Reference saved memories also turns off chat history What it learned from past chats is scheduled for deletion within 30 days; the chats stay

Memory can also draw on your custom instructions, files in your Library and connected apps such as Gmail, depending on plan and region. It keeps what ChatGPT judges relevant, which isn’t every detail. When a reply draws on your memory, Sources may appear under it, pointing to a past chat, saved memory or other source that contributed.

Since June 2026, OpenAI has been rolling out improved memory, which keeps a memory summary up to date on its own. You can read and correct that summary, or, where the option is offered, switch back to the older list of saved memories, in Settings > Personalization > Memory.

Which ChatGPT plans have memory?

  • Free: memory and memory with past chats, both marked “Limited” on OpenAI’s pricing page.
  • Go: yes. Plus and Pro: “Expanded”.
  • Business, Enterprise and Edu: available when the workspace allows it. Workspace owners and admins decide whether memory is on for the workspace, and each member can still turn their own off.

OpenAI adds that memory features “can vary by plan, region, platform, and workspace settings”, so a setting described here may not show on your account yet.

Does ChatGPT remember chats inside a project?

Yes, and you choose how far that memory reaches. Each project uses one of two settings, under the project’s ••• menu, Project settings, Memory. You can change it later; a change may take a few hours to apply.

  • Default memory. Your saved memories apply in the project, and chats can draw on the project’s other chats. On Enterprise and Edu, chats stay inside the project. On other plans, Business included, project chats can also reference chats outside the project, and be referenced by them.
  • Project-only memory. Chats in the project draw on each other and on nothing outside it. Your saved memories aren’t used there, and chats outside the project can’t see in. For client work, this is the setting that keeps one client’s details out of another’s chats; more in Why ChatGPT mixes up your chats.

Shared projects always use project-only memory. There’s no list of what a project remembers; to stop one chat informing the project, delete it or move it to another project.

How do you make ChatGPT remember something?

  1. Check memory is on: Settings > Personalization > Memory.
  2. Tell it in a chat, for example: “Remember that the client’s brand name is always written in capitals.” ChatGPT saves it to memory.
  3. For rules that should apply to every reply, write them in custom instructions (Settings > Personalization).
  4. For one body of work, make a project and add its files and instructions, so every chat in it starts from them.
  5. To check, ask “What do you remember about me?” or open the memory summary.

What does “memory full” mean?

In an October 2025 release note, OpenAI describes “memory full” as the state your account reaches when saved memories hit capacity. The same note brought automatic memory management to avoid it, first for Plus and Pro on the web: ChatGPT keeps the most relevant memories prioritised and moves less important ones to the background. OpenAI’s current memory page doesn’t mention “memory full” or a capacity for saved memories. It says saved-memory controls can include search, sorting, automatic prioritisation and version history, depending on plan and platform, and that what ChatGPT references from chat history has no separate storage limit. If your account still shows “memory full”, delete the saved memories you no longer need.

How do you turn ChatGPT memory off, or chat without it?

  • Turn it off: Settings > Personalization > Memory. Turning memory off doesn’t delete your past chats.
  • Delete what it knows: from the memory summary’s ••• menu, Delete and turn off memory removes what the summary shows and turns memory off.
  • One chat without it: start a Temporary chat and choose Unpersonalized before your first message (Personalized uses your existing memories). Neither kind creates or updates memories, and a temporary chat stays out of your history unless you save it. OpenAI may keep a copy for up to 30 days for safety.
  • Deleting a chat doesn’t necessarily delete a saved memory made from it, so delete the memory as well. OpenAI lists archived chats, alongside regular ones, among the places to clear when you want something forgotten.

Does ChatGPT share memory with your team?

No. OpenAI’s business FAQ says memories are “tied to each individual account” and “not transferable to other users, even within the same Business workspace”.

Project sharing is on Free, Go, Plus, Pro, Business, Enterprise and Edu. Members with chat access see the project’s chats, files and instructions, and ChatGPT draws on anything in the shared project, other members’ chats included. Shared projects run on project-only memory and “cannot access an individual member’s memories or context from outside the project”.

The same line holds when you share a chat or a Page. A shared chat link leaves out memory sources, and OpenAI says of its new Pages: “Sharing a Page does not give other people access to your private chats or memory”.

So inside ChatGPT, a shared project is where a team keeps its files, instructions and chats together, and everyone in it has to be in ChatGPT.

Can Claude or another AI see your ChatGPT memory?

No. It stays in your ChatGPT account. The nearest thing to a bridge is a one-off copy: Claude’s memory import gives you a prompt to paste into ChatGPT, and you paste ChatGPT’s answer into Claude. That copies what ChatGPT chose to list about you on that day. Nothing updates after that, and your colleagues’ memories and chats never come along. The full move is in How to move from ChatGPT to Claude without losing your projects, and Claude’s side is in Does Claude remember across chats and projects?

Say your strategist works in ChatGPT and your designer in Claude. Each AI remembers its own person well. Neither knows what the other was told.

Where Alcoven fits

Memory is the right tool for remembering you. A team split between ChatGPT and Claude needs something more: one project both AIs open, where the brief, the documents and the tasks are the same for everyone.

That is what Alcoven keeps. You add Alcoven to ChatGPT, on plans where an MCP tool can write, or to Claude as a connector, and sign in once to reach every project you’re on; from Claude Code or Codex, you clone the project with git. From then on your AI reads the project and writes back to it, and each change arrives under your name for everyone else on the project. Alcoven runs no model of its own for this. Documents keep every past version, so anyone on the project can open one and see who changed what since they last read it. Your ChatGPT memory carries on as before, beside it.

Free for one project and up to five people writing to it, no card. Start a project free

Sources

All checked on 30 September 2026.