ChatGPT starts remembering users across conversations

Every chat used to start from zero

A conversation with a chatbot is normally a self-contained episode. When the window is closed, the context built up inside it does not follow the user into the next session; background, preferences and working habits have to be explained again. Memory was OpenAI's answer to that: a way for ChatGPT to carry some information about a user from one conversation to the next instead of starting blank each time.

A small test in February 2024

On 13 February 2024, OpenAI began testing memory with a small portion of free and Plus ChatGPT users ^[1]. The mechanism was described simply: ChatGPT could remember information about users across conversations ^[1], so that something a user had mentioned once would still be there later, without being repeated.

Opening up to Plus subscribers

On 29 April 2024, memory became available to all Plus subscribers, except in Europe and Korea, and it could be turned on or off in settings ^[2]. Handing the switch to the user was a central part of the design: if someone did not want a topic carried forward, they could stop it.

Controls, and a chat that leaves no trace

The settings did not stop at a single switch. Users could turn off referencing saved memories or chat history at any time in settings ^[4]. OpenAI also introduced temporary chat, a mode for conversations without memory ^[4], and said that GPTs would have memory too ^[4]. The result was a set of choices rather than one setting: everyday use in which context persists, and a separate window for conversations the user would rather not have remembered.

Reaching every major tier

On 5 September 2024, memory reached Free, Plus, Team and Enterprise users ^[2]. For Team and Enterprise customers, OpenAI offered the feature as a way to work more efficiently ^[4] — a project's background, for instance, does not have to be restated at the start of every session.

A capability that kept growing

By April 2025, ChatGPT had gained the ability to remember past conversations ^[3], extending what began as a limited test to the wider history of a user's chats.

Why it mattered

Memory changes the shape of the interaction more than it changes the model itself. An assistant that remembers is used differently from one that forgets: a user can pick up a thread where it was left off, instead of restating the premises each time. On the other hand, remembering also makes retention something the user has to see and manage. Because memories persist, people need to know what is being kept and how to stop it being referenced — which is why the off switches and temporary chat arrived together with the feature ^[2][4]. Memory therefore sits at the meeting point of two questions that conversational products keep running into: how much context should carry over, and who decides.