Have you ever had this small, strange feeling with ChatGPT? You open a new chat, and it already seems to know you a little. Maybe your writing style. Maybe that you are vegetarian. Maybe that you are planning a trip in July. As of June 2026, OpenAI is rolling out a newer memory system for ChatGPT. It starts with Plus and Pro users in the US, and it is designed to stay more up to date by building a memory summary from past chats, not only from facts you clearly said, “Please remember.” (help.openai.com)
Now picture this. On Monday, you ask for easy lunch ideas and mention that you do not eat meat. On Friday, you ask again, and ChatGPT suggests a vegetarian lunch without making you repeat yourself. That is the bright side. OpenAI says memory can use saved memories and chat history, and the newer system updates memories automatically so old notes do not become stale so easily. You can even ask, “What do you remember about me?” and check the memory summary. (openai.com)
But here is the turn. Deleting one chat may not fully erase what ChatGPT knows. OpenAI says saved memories are stored separately, so to fully remove something, you may need to delete the memory itself and also delete the chat, files, or connected apps where that information appeared. OpenAI also warns that sensitive information may appear in memory if you share it. And if “Improve the model for everyone” is on, content from chats and memories may be used to improve models. (help.openai.com)
So the real question is not only, “How much can AI remember?” It is also, “What do I want it to remember?” Memory can make AI feel less like a tool and more like a helper. But a good helper still needs clear boundaries. In the age of smart memory, forgetting may become a skill too. (help.openai.com)










