Context that survives the end of a chat
AI memory stores your preferences, working assumptions, and past conversations, then applies them in later chats. Tell an assistant that you want concise writing, or that you work on a Mac, and it can reflect that without being told again.
ChatGPT and other conversational assistants offer memory in various forms. What is stored, how long it is kept, which plans include it, and how to delete it all differ by service.
How it differs from chat history and context
Remembering what you said earlier in the same chat is the conversation’s context. Carrying preferences and facts into a brand-new chat is what is usually called memory.
ChatGPT separates saved memories, which you add explicitly, from references to your chat history. Project memory narrows the scope further, drawing only on the conversations and files inside one project.
The context window is the ceiling on how much information a model can read in a single pass. Memory is best understood as a mechanism that selects relevant stored information and adds it to the context of a new conversation.
Memory is not model training
Something being saved to memory is not the same as your input being used to train the model. Memory shapes the answers you personally receive; whether your data is used for service improvement is governed by separate settings and terms.
Turning memory off does not necessarily delete everything at once. Saved memories, the original chats, and uploaded files may be managed separately, so check each one when you want something fully removed.
Using it safely
- Review what the assistant has remembered from time to time
- Delete outdated information and personal details you no longer want stored
- Use a temporary chat for sensitive or one-off questions
- Keep work, personal, and per-project memory separate
Memory saves you from repeating yourself, and it gets more useful the longer you use it. The trade-off is that deciding what should and should not be remembered stays your responsibility.