Google's latest AI feature, Personal Intelligence, is now rolling out to subscribers of Google AI Pro and AI Ultra, and it's already demonstrating an unsettling ability to surface deeply personal information from users' digital footprints. In a hands-on experiment, tech journalist Pranav Dixit found that the AI could retrieve his license plate number and even infer his parents' vacation preferences from a mix of emails, photos, and search history.
Dixit, writing for Business Insider, described the experience as akin to Google having "quietly been taking notes on my entire life and finally decided to hand me the notebook." The feature, which is integrated into the Gemini app and Google's search AI Mode, can scour Gmail and Google Photos, and a more powerful version released earlier this month also taps into Search and YouTube history.
The rollout marks a strategic move by Google to differentiate its AI offerings from competitors like OpenAI. With decades of user data from billions of people, Google can infer a remarkable amount from search queries alone, and Gmail accounts are often filled with confirmations for appointments, bookings, and purchases.
For privacy advocates, the feature raises red flags. Google, however, maintains that it is handling personal data carefully. Josh Woodward, a Google VP, wrote in a recent blog post that the AI is trained on user prompts and responses, not on photos or emails. "We don't train our systems to learn your license plate number," Woodward said. "We train them to understand that when you ask for one, we can locate it."
Dixit's testing revealed both the utility and the potential pitfalls. When he asked for sightseeing ideas for his parents, the AI correctly deduced they had already done many hikes in the Bay Area and suggested museums and gardens instead. It pieced this together from "breadcrumbs" such as emails, photos of a forest, a parking reservation, and a search for "easy hikes for seniors." It also identified his license plate from photos and reported his car insurance renewal date from email scans.
Why This Matters for AI Trust
The feature's ability to make AI seem more humanlike is a double-edged sword. Dixit noted that while the assistant feels genuinely useful, it also risks fostering an illusion of intimacy. He lamented that he could "pour my soul into ChatGPT and get a smart answer," only for it to "forget I existed like a genius goldfish." Experts have previously raised concerns about AI memory features making chatbots appear too lifelike, potentially leading users to form unhealthy attachments.
As Google expands Personal Intelligence, users must weigh the convenience of a hyper-personalized assistant against the implications of an AI that knows where you've been, what you've bought, and who you've traveled with. The feature is currently available only to paying subscribers, but its broader rollout could reshape how people interact with their data—and how much they're willing to share.