We’re testing memory with ChatGPT to make your experience more seamless by saving important details across chats, so you won’t have to repeat yourself. This feature helps make future conversations more helpful.
You’re fully in control of ChatGPT’s memory. You can ask it to remember something, view what it recalls, and even delete specific memories either conversationally or through settings. Memory can also be turned off completely.
This week, we’re rolling out memory to a small group of free and Plus users to gather feedback. Broader rollout plans will be shared soon.
How Memory Works
As you interact with ChatGPT, it can remember key details from your conversations, improving the quality of future responses. For instance:
- If you mention you prefer meeting notes with bullet points and action items, ChatGPT will structure future summaries that way.
- If you share that you own a neighborhood coffee shop, ChatGPT will remember this when helping craft messaging for a new location.
- If you tell ChatGPT your child loves jellyfish, it will suggest incorporating that into a birthday card design.
You’re In Control
You can turn memory off at any time (Settings > Personalization > Memory). With memory off, ChatGPT won’t store or use any memories.
To delete specific memories, simply ask ChatGPT to forget or manage them in settings. Memory works across interactions, meaning deleting a chat doesn’t erase its associated memory—you’ll need to delete the memory itself.
ChatGPT may use the content you provide, including memories, to improve its models for everyone, unless you opt out through Data Controls. Note that content from Team and Enterprise accounts won’t be used to train models.
Temporary Chat for No Memory
If you’d prefer a conversation without memory, use temporary chat. These conversations won’t appear in history, won’t store memories, and won’t contribute to model training.
Custom Instructions and Memory
Custom Instructions let you guide ChatGPT on how to respond, while memory captures information shared in conversations. This combination allows ChatGPT to become more personalized and responsive over time.
Privacy and Safety Standards
We’re evolving our privacy and safety protocols to address memory’s impact. ChatGPT is designed to avoid remembering sensitive information, like health data, unless explicitly requested.
Memory for Team and Enterprise Users
For Team and Enterprise users, memory helps increase efficiency by learning individual preferences and reducing the need for repetitive instructions. For example, ChatGPT can remember your preferred tone and structure for content or your preferred coding languages for programming tasks.
Memory in Team and Enterprise accounts remains secure and excluded from model training, with full control over how and when memories are used. Account owners can disable memory for the organization at any time.
Memory for GPTs
GPTs, too, will have distinct memories. Builders can choose to enable memory, and each GPT will store its own memories. For example, a book recommendation GPT can remember your favorite genres for tailored suggestions.
To interact with memory-enabled GPTs, you’ll need memory on. Each GPT will have its own separate memory, so details shared with ChatGPT won’t carry over unless re-entered.
Memory is now available to ChatGPT Free, Plus, Team, and Enterprise users. Based on user feedback, ChatGPT will notify you when a memory is updated, and you can easily review or delete those updates by accessing the “Manage memories” option in settings.