ChatGPT's Automated Memory System Now Builds User Profiles Without Manual Input

OpenAI updated ChatGPT's memory system in June, shifting the tool from a manual command-based approach to an automated process that continuously builds a profile of each user. According to a report by Complete AI Training published on 21 September 2026, the change affects every response the chatbot delivers and carries particular implications for professionals in fields such as marketing, legal, healthcare, and customer support.

From Manual Commands to Automated Profiling

When memory first launched in April 2024, users had to explicitly instruct the chatbot to retain information. Those stored entries were static and did not update over time. The June update introduced what the source describes as "flexible memories that update over time," meaning the system can, for example, move a noted event from future to past without any user correction.

OpenAI describes three goals for the updated feature: carrying forward useful context, following preferences and constraints, and staying current as facts change. The trade-off, according to the source, is reduced user control — the chatbot now independently decides what to save from conversations without requiring user approval.

Risk of Incorrect Inferences

The source notes that the automated system can produce inaccurate inferences. As illustrative examples, the report states that spending a week researching a vacation destination could lead ChatGPT to assume the user lives there, while researching home repair could cause it to tag the user as a DIY expert. For business users, the report warns that a flawed memory profile can quietly steer outputs in unintended directions — producing location-specific suggestions based on client research, or skipping explanatory steps by assuming expertise the user does not have.

How Users Can Review and Edit Memory

The report outlines several user controls. On the web app, the memory profile is accessible via the profile icon, then Personalization, then the Memory section. A toggle switch allows the feature to be disabled entirely; turning it off stops new memories from forming but preserves existing ones. Users can select "Manage" next to Memory summaries to review stored interests, preferences, and communication preferences, and can correct or remove individual entries. To delete the entire profile, users can select "Delete and turn off memory" from the three-dot menu. If memory is re-enabled after deletion, the report states that "ChatGPT will re-index those chats and rebuild its understanding of you from scratch."

A separate temporary chat feature is also available. The source notes that nothing typed in a temporary chat is committed to memory, and the conversation disappears when closed — described as useful for sensitive queries or one-off research.

Implications for Content and Business Teams

The source characterises memory profile management as a practical responsibility for content teams and business professionals, comparing it to reviewing browser cookies or app permissions. The report suggests a regular audit after project completion to prevent what it describes as "cross-contamination between work streams." It notes that "the responsibility for accuracy still sits with the user."