Web assistants and the agents that act for their users keep their own memory: after each conversation, the model writes notes about the user that personalize its later answers and actions. In ChatGPT, most memory entries are written by the system rather than by the user (Dash et al., 2026). The notes are a user model, and data-protection law lets users ask a service to correct or erase what it holds about them. The diagram follows one note through that loop; each step carries the paper's measured result.