AI memory (chat memory)

It also means the assistant keeps details you may have forgotten you shared, which is why memory belongs on your privacy checklist next to the training opt-out.

How it works

Each product handles memory differently. ChatGPT draws on saved memories (details you asked it to keep, or that it judged useful) and on your past chat history. OpenAI notes that saved memories are stored separately from chats, so deleting a chat does not delete a memory taken from it. Claude summarizes your past chats into a memory summary you can view and edit in settings, and keeps a separate memory for each project.

Both have a way to chat without memory: ChatGPT's temporary chats and Claude's incognito chats do not create or update memories. Memory is also different from the context window, which only covers the current conversation.

Example

A bakery owner with two shops tells her assistant to remember the locations and that she prefers short answers. Weeks later she asks for a staff rota template and gets one built for two sites, with no setup. During a review in the spring she opens the memory settings and deletes an old entry about a flour supplier she dropped.

Common mistake

Never looking at what has been stored. Memory collects stale details, like an old job or a finished project, and sensitive ones, like a health question or a client's name, and the stale ones can skew later answers. Open the memory settings every few months, delete what is wrong or private, and use a temporary or incognito chat for one-off sensitive questions.