Imagine: you deleted an email from Gmail — but Gemini still "remembers" its content for several days. This is not a bug. This is documented behavior of the new "Personalized Intelligence" feature, which is now available to Ukrainian users.
What exactly connects — and how
"Personalized Intelligence" allows Gemini to read your Gmail, Google Photos, YouTube history, and search history to provide contextual answers. Questions like "did my package arrive?" or "find photos from last year's birthday" — Gemini now searches for answers not on the internet, but in your personal data.
In the United States, the feature launched as opt-out — meaning Gmail access was enabled by default. As char.com notes, in Europe and GDPR regions, Google was forced to make it opt-in: connecting requires explicit consent. For Ukraine, this variant currently applies — connection is voluntary.
Where there is a gap between "disabled" and "deleted"
This is where it gets interesting. Google's official documentation directly warns: if you delete or update data in a connected application — for example, an email from Gmail or a photo — Gemini may not react to this for several days. In other words, there is a window between your action and its actual reflection in the AI's behavior.
Separately — the question of conversations being reviewed by people. As heydata.eu reports, even if you disable activity saving, conversations can be stored for up to 72 hours in Google's operating systems. The company acknowledges that human reviewers may analyze interactions to improve AI quality — and recommends not entering anything you would not want to show to a live person.
"Gemini does not learn directly from your personal data"
— Google blog on Personal Intelligence
The wording is precise but narrow: "does not learn directly" does not mean "does not store" and does not exclude the use of aggregated or anonymized data in the future.
Why this matters beyond the paranoid
The competitive context sets priorities. As Fortune writes, Microsoft is expanding Copilot with long-term memory and integration with Google Drive and Gmail, and Anthropic launched Claude Cowork — an agent that works with user files. All major AI labs are moving in one direction: toward models that know as much about you as possible. The difference is in which data, how long, and with what level of user control.
Google has a structural advantage in this race: billions of people already store emails, photos, routes, and search queries in its ecosystem. "Personalized Intelligence" is monetization of the context that already exists. The question is not whether to trust Google in general, but whether the risks change when this context becomes an active tool of an AI agent.
How to manage access
- Connecting each service (Gmail, Photos, YouTube) is separate and voluntary
- Disable learning and human review: myaccount.google.com → Data and Privacy → Activity in Gemini → Disable
- After disabling activity, future conversations will not be used for training — but 72-hour operational storage remains
- Deleted data in connected applications syncs with a delay — up to several days
The feature is available for Google AI Pro and Ultra subscribers. For free accounts — later, timelines have not been announced.
The real test for "Personalized Intelligence" will come not at launch, but when the first documented case of a leak or unauthorized access through connected accounts appears: if Google responds with transparent communication and a quick patch — trust in the feature will grow; if with silence — mass disconnections will follow.
