What did OpenAI change in ChatGPT memory?

OpenAI upgrades ChatGPT memory and adds user controls

OpenAI has rolled out an update to ChatGPT’s “memory” system, describing it as both more capable and more compute-efficient. Alongside the underlying memory changes, OpenAI is introducing a summary page that lets users see what the assistant remembers and steer that behavior over time.

The key shift is that memory is no longer just an invisible capability: users will be able to review the stored information and influence how it’s used. That matters because memory systems can affect personalization, long-running context, and potentially privacy expectations—especially for users who previously couldn’t easily verify what was being retained.

OpenAI’s packaging of these improvements also signals a broader product direction: treating memory as a user-facing feature rather than a backend-only mechanism. The practical implication is that people can be more intentional about what the system should learn, and they can correct or remove information once they can identify it.

In day-to-day usage, a more efficient memory architecture may also reduce latency or resource usage compared with prior implementations, which can translate into smoother interactions, particularly for frequent users.

If you’re testing ChatGPT’s personalization today, this update changes the feedback loop: instead of assuming the model “just remembers” correctly, you’ll be able to inspect the memory contents and adjust them.

Overall, the announcement frames the change as both a capability upgrade and a transparency/control improvement—two factors that are increasingly important as conversational AI becomes more persistent.