ChatGPT’s memory can change an answer before a user types any new background because the system may already be drawing on preferences, projects and constraints collected across earlier conversations.
That is the premise behind WIRED’s guide to understanding what ChatGPT thinks it knows about a user. The practical question is not just whether memory is on, but which sources are available and how the remembered context is reviewed, corrected or removed.
“Dreaming” updates a broader picture over time
OpenAI’s improved memory architecture uses a background process it calls “dreaming” to synthesize context across conversations. Unlike a fixed note such as “I’m traveling to Singapore in July,” the system is designed to recognize that the trip later ended and stop treating the user as though they are still there.
Depending on the account, region and workspace, OpenAI’s current help documentation says memory can draw on past chats, saved memories, custom instructions, Library files and connected apps. ChatGPT does not retain every detail from every conversation; it selects context it considers relevant to the current response.
Those choices can affect recommendations, explanations and web searches. A dietary preference may narrow a restaurant search, while an ongoing project can change which constraints the assistant foregrounds. The benefit is continuity, but stale or mistaken context can also steer an answer in the wrong direction.
The memory summary is a control surface, not a full audit log
When available, the memory summary gives users a high-level view of information ChatGPT may use for personalization. Users can enter corrections, edit remembered information or ask that a detail not be mentioned again. Personalized responses may also show Sources pointing to a past chat, saved memory, file, custom instruction or connected email.
OpenAI cautions that the summary and Sources do not necessarily expose every factor that shaped a response. The controls also vary by plan, platform, region and workspace, so two users may not see identical menus.
Turning memory off and deleting data are different actions
Turning Memory off stops the personalization feature but does not delete past chats. Deleting a chat may still leave a separate saved memory created from that conversation. OpenAI says fully removing a detail can require deleting both the saved memory and the chat where it appeared, plus any relevant files or connected-app source.
Temporary Chat offers a cleaner boundary: it can be used without creating or updating memories. On personal accounts, remembered information may also be used to improve OpenAI’s models when “Improve the model for everyone” is enabled. Business, Enterprise, Edu and Healthcare workspace content is not used for training by default.
