Gemini, Google's AI chatbot, remembers details from past conversations and uses them to tailor later replies. A new feature, Memory Summary, lets you see what it has stored about you in one place.
What you can do with it
You can correct wrong information, delete details that are outdated or unnecessary, and add new ones — a changed workplace, say, or a new preference.
The stored memory is organised into categories: identity, relationships, professional responsibilities, interests and preferences.
You can give Gemini instructions in plain language to add, correct or remove memories, and if an instruction is unclear it asks for more detail. A pencil icon allows direct editing. A Memory History option, under the three-dot menu, shows when memories were added or changed.
Where to find it
In Gemini's web interface: Settings → Personal Intelligence → Memory Summary. You can also simply ask Gemini "What do you know about me?" to review what it holds.
Who gets it
It is initially available only to Google AI plan subscribers in the United States. It requires a personal Google account with the Keep Activity and Memory settings turned on.
That last condition is worth reading twice: the feature that lets you inspect and delete stored memory only works if you have memory switched on in the first place. Turning memory off remains the stronger privacy choice; Memory Summary is for people who want the personalisation and want to see its cost.
Why the history view is the best part
Of everything announced, Memory History is the feature that does real work.
A list of what a system currently knows about you is useful but incomplete — it cannot tell you when something was inferred, or from which conversation. A change log can. That matters because the uncomfortable entries in an AI's memory are rarely things you stated outright; they are things it concluded. Knowing an assistant decided you had changed jobs, and roughly when, is the difference between auditing a system and simply reading its summary of you.
What to check when it reaches here
When this arrives in Bangladesh — and on the current pattern that will be months after the US launch — three categories are worth looking at first.
Relationships, because names of family members and colleagues accumulate from ordinary questions and are the most sensitive thing in the list. Professional responsibilities, because an assistant that has inferred your employer and role from work questions is holding information your employer may consider theirs. And identity, because that is where guesses about age, location and language settle, and where a wrong inference quietly shapes every answer you get afterwards.
Deleting a memory is not the same as undoing it
One limit is worth understanding before anyone treats this as a full reset. Removing an entry from Memory Summary stops it being used in future replies. It does not reach back into the conversations that produced it, and it does not necessarily remove the underlying activity log that the memory was drawn from — those are governed by the separate Keep Activity setting.
So the page is best thought of as a control over what the assistant acts on, not a record of everything held about you. That is still a real and useful power. It is simply a narrower one than the word "delete" implies, and the distinction is the kind that only becomes obvious when someone assumes otherwise.
