ChatGPT's Memory Is Shaping Its Answers-And You Can Edit What It Thinks It Knows

AI chatbots now remember your conversations by default, storing past interactions to give context to future answers. Tell it you're vegetarian, and it won't suggest meat recipes. But that memory can also skew responses in ways you didn't intend. Spend a week researching light-switch repairs and ChatGPT may peg you as a DIY devotee; look up spots near a vacation destination and it might assume you live there.

When ChatGPT launched memory in April 2024, you had to explicitly tell it to remember something, and those memories stayed frozen-an upcoming party remained upcoming even after the date passed. A June update introduced what OpenAI calls dreaming: memories are now flexible, built and edited over time, and saved automatically without manual requests. The chatbot decides what's worth remembering, and it won't always judge correctly. OpenAI says memory aims to carry forward useful context, follow preferences and constraints, and stay current.

To see what ChatGPT has stored, open the web app, click your profile icon, choose Personalization, and scroll to Memory. On mobile, tap the menu, then your profile icon, then Memory. A toggle lets you disable memory, though existing memories remain stored. Under Memory summaries, choose Manage to review the AI's profile of you-covering interests and answer preferences-and select Make a correction or Don't mention this again. A prompt box lets you issue commands like "forget that I'm a journalist." To erase everything, pick the three dots, choose Delete and turn off memory. Past conversations stay put, and if you re-enable memory, they'll be indexed again. Temporary chats, started via the dotted speech bubble on the web, are never committed to memory.