ChatGPT's memory feature: what it stores and how to control it
Original: What ChatGPT Thinks It Knows About You Is Affecting Its Answers. Here’s How to Change That
Why This Matters
Automatic AI memory raises real personalization vs. privacy tensions as chatbot use scales across daily workflows.
OpenAI's ChatGPT now automatically builds and updates a memory profile of each user — a system called 'dreaming' introduced in June 2026. These stored assumptions shape every response, but users can view and edit what the chatbot thinks it knows about them via the Personalization settings.
ChatGPT's memory has evolved well beyond its April 2024 debut, when users had to manually tell the bot what to retain. That early system was rigid: if you mentioned an upcoming party, ChatGPT would treat it as future indefinitely, even after it passed.
The June 2026 'dreaming' update changed that. Now, ChatGPT continuously and automatically builds a flexible memory profile — inferring preferences, updating context over time, and deciding on its own what's worth remembering. OpenAI frames the three goals as: carrying forward useful context, following user preferences and constraints, and staying current.
The trade-off is control. Because the bot makes its own judgments, it can draw wrong conclusions — labeling someone a DIY enthusiast after one week of light-switch research, or assuming a vacation destination is a home city. These false assumptions then color every subsequent answer.
Users can audit and edit what ChatGPT has stored. On the web app: profile icon → Personalization → Memory. On mobile: menu → profile icon → Memory. Memories are shown as summaries and can be deleted individually or cleared entirely. Users can also disable the memory feature altogether through the same settings panel.