What is ChatGPT memory?
ChatGPT memory is the set of features that lets ChatGPT carry details from earlier conversations into new ones, so you don't have to repeat your job, your projects or your preferences. It has two parts: saved memories, specific details ChatGPT keeps (often because you asked it to), and reference chat history, where ChatGPT draws on relevant past conversations to personalize answers.
Both are managed from Settings > Personalization > Memory. OpenAI's ChatGPT memory documentation points to Settings > Personalization for these controls. The full details are in OpenAI's help article Memory in ChatGPT.
Memory is separate from custom instructions. Custom instructions are text you write and control word for word. Memory is what ChatGPT picks up from your chats, and its contents change over time.
What is the difference between saved memories and reference chat history?
| Saved memories | Reference chat history | |
|---|---|---|
| What it is | Specific details ChatGPT stores, such as "prefers metric units" | Relevant context drawn from your past conversations |
| How it gets there | You ask ChatGPT to remember, or it saves a detail it judges useful | ChatGPT uses past chats as needed |
| Can you see each item? | Yes, in Manage memories or Saved memories | Not item by item; the memory summary is not exhaustive |
| Turning it off | Switch off Reference saved memories (also turns off chat history) | Switch off Reference chat history only |
| Deleting a chat | Doesn't delete a saved memory from that chat | Removes that chat from what can be referenced |
The key practical point from the help center: saved memories live apart from your chats. Deleting a conversation does not delete a saved memory that came out of it, and deleting a memory does not erase the detail from the original conversation. To remove something completely, delete both.
Reference chat history is less fixed. In June 2026 OpenAI announced that memories are updated automatically, with ChatGPT keeping track of the details it judges most important, as part of a newer memory system it calls "dreaming."
How do you view and edit what ChatGPT remembers?
Go to Settings > Personalization > Memory. Depending on your account's memory experience, you will see some of these controls: Memory, Reference saved memories, Reference chat history, Memory summary, Manage, or Saved memories.
- Memory summary: an overview of what ChatGPT has learned. OpenAI notes it is not a complete list, because memory comes from a continually updated synthesis of your past chats.
- Manage memories or Saved memories: the list of individual saved memories. Delete any item you no longer want.
- In a chat: ask "What do you remember about me?", or say "Forget that I work at Acme" to remove a detail.
Since May 2026, a Sources icon under responses shows which saved memories, past chats or custom instructions influenced that reply. OpenAI's release notes say memory sources are available on all ChatGPT consumer plans. It is the fastest way to find a stale memory: when an answer assumes something out of date, open Sources and edit or delete what it used.
How do you turn ChatGPT memory off?
You have three levels of control:
- Everything off: switch off Reference saved memories. This also turns off Reference chat history.
- Chat history only: switch off Reference chat history and keep saved memories on. OpenAI's help center says the information ChatGPT remembered from past chats is then scheduled for deletion within 30 days, while the chats themselves stay in your history.
- One conversation: use a temporary chat (next section).
Turning memory off does not delete your chats or your custom instructions. If you want a clean slate, delete the saved memories you no longer want, then decide whether to keep chat history referencing on.
What are temporary chats?
A temporary chat is a one-off conversation that ChatGPT doesn't personalize or remember. Per OpenAI's help center, including Data controls in ChatGPT, temporary chats:
- don't create or update memories, and by default don't use memory or custom instructions (you can choose a personalized temporary chat when you start it),
- aren't saved to your chat history,
- aren't used to train OpenAI's models,
- may be kept by OpenAI for up to 30 days, according to its help center.
If you save a temporary chat, OpenAI's retention article says it becomes a regular chat and follows the rules for saved chats. Use temporary chats for questions you don't want shaping future answers, for testing a prompt without your preferences mixed in, or for sharing a screen without your personal context showing up.
How does memory differ by plan?
As of October 2026, OpenAI's pricing page shows memory growing with the plan, with Free the most limited and paid plans offering more. OpenAI does not publish exact memory sizes, so check the pricing page for your plan's current label.
The timeline matters for Free users. The ChatGPT release notes say memory improvements that reference recent conversations started rolling out to Free users on June 3, 2025. In June 2026, OpenAI launched its newer "dreaming" memory system for Plus and Pro users in the US, said it would roll out to Free and Go users and more countries over the following weeks, and gave Plus and Pro twice as much memory capacity. Region can also matter: in 2025, OpenAI made the chat history upgrade opt-in for Free users in the EEA, UK and Switzerland rather than automatic.
Business, Enterprise and Edu workspaces follow their own memory settings, which admins control. OpenAI keeps a separate business version of its memory FAQ for those plans.
How does memory work with projects and GPTs?
Projects have their own memory setting. With project-only memory, saved memories from outside the project are not used in its chats, though chats can reference other conversations in the same project. Shared projects always use project-only memory, so a teammate never sees answers shaped by your personal memories. See the ChatGPT Projects guide for the details.
GPTs start fresh. OpenAI's help center says a GPT doesn't use your saved memory, custom instructions or past conversations, so each GPT chat begins without your personal context.
Memory vs instructions: what should go where?
Memory is convenient, but it's not a contract. ChatGPT decides what to keep, what it keeps can shift, and the summary you see is not complete. OpenAI's own docs say to treat memories "as a helpful recall layer, not as the only source for rules that must always apply."
A simple rule:
- Must always apply (format, tone, your role, hard rules): write it in custom instructions or project instructions.
- Nice to know (what you're working on this month, tools you use): let memory handle it.
- Sensitive or temporary: use a temporary chat, or delete the memory afterward.
Claude handles this differently, with per-project memory and incognito chats. The Claude memory guide covers how it compares.
Keep the instructions that matter in a file
Because memory changes on its own, the prompts and instructions you rely on are worth keeping as text you control. Kipt stores prompts as Markdown in a private GitHub repo you own, with Cmd+K search and blanks like {{audience:developers}} that you fill in when you copy one into ChatGPT, a project or any other tool. That way your best instructions don't depend on what a memory system decided to keep.