Personalization should not require people to accept an invisible, permanent profile. Memory should be something users can inspect and negotiate.
Personalization is a continuing inference about a person, not a one-time event. A conversation can include a joke, a temporary plan or information about someone else. I would therefore treat memory as something the user can inspect and renegotiate, with special care before sensitive inferences become durable context.
A small assumption can spread
Suppose an assistant remembers that a user is looking for a new job after helping with a friend’s résumé. That assumption may later shape career advice or a workplace email. A visible source for the memory would make the error easier to identify; a generic personalization switch would not explain it.
Deletion should match the user’s mental model
OpenAI’s Memory FAQ explains that underlying sources may need to be removed as part of deleting information. This illustrates the gap between a user’s simple intention to forget and the product’s multiple stores of context. Our proposal is for the interface to guide that process and clearly state its limits.
Source: OpenAI · ChatGPT Memory FAQRequiring a decision for every remembered detail could make a helpful feature exhausting. It may also burden people who rely on continuity for accessibility or who find repeated configuration difficult. A system that remembers nothing until explicitly instructed can still frustrate user intent.
Where the debate remains open
A reasonable compromise is to distinguish explicit preferences from uncertain inferences, provide a concise review surface and let temporary context expire. Sensitive details deserve more deliberate treatment. The specific defaults should be tested with users whose needs differ, not selected solely for engagement.
What would change this view?
I would revise the proposed review pattern if users could not understand it or if it failed to correct consequential mistaken assumptions. The evidence should include what people thought was remembered and deleted, compared with what the product actually did.
For more reading
Background evidence for this editorial argument, including the limits and counterpoints. The conclusions are the site’s interpretation.
- ChatGPT Memory FAQ
Explains memory controls and deletion. Check which experience and settings apply to your account.
- Gemini Apps Privacy Hub
Details activity, connected apps, retention and human review for Gemini Apps; other deployments can differ.
- Guidance on AI and children · version 3
Connects safety, privacy, inclusion, development and children’s participation in AI design.
Sources reviewed 13 September 2026. Product documentation can change. How we use evidence
How does this argument land with you?
Participate anonymously. No account required.