Google taps its massive data advantage with new Gemini feature

Google is connecting its AI assistant Gemini with Gmail, Google Photos, YouTube, and search. The new feature gives the company a lead that OpenAI and Anthropic will struggle to match.

Google has launched a feature called "Personal Intelligence" for Gemini. It connects the AI assistant with Gmail, Google Photos, YouTube, and Google Search, feeding that information directly into chatbot responses. This gives Gemini access to personal data that no other AI provider has.

Personal Intelligence is available as a beta for paying subscribers in the US. Workspace users in business, enterprise, or education tiers are excluded for now. Google plans to expand the feature to other countries and the free tier later.

Gemini can now pull license plates from photos and receipts from emails

The system combines information from multiple sources: text, photos, videos, emails, and search histories. It can make product recommendations based on shopping receipts or plan trips based on past vacations.

Josh Woodward, VP of the Gemini app, Google Labs, and AI Studio, shows how this works: When shopping for tires for his minivan, he asked Gemini for the tire size. The chatbot provided the specs and suggested options. It referenced family trips it found in Google Photos.

When Woodward needed his license plate number, Gemini pulled the seven-digit number from a photo. The assistant figured out the exact vehicle trim by searching through Gmail.

Two decades of user data become a competitive moat

Google isn't shy about leveraging the data advantage it has built over the last two decades. The company sits on billions of emails, photos, and search histories; first-party data that neither ChatGPT nor Claude can access at a comparable scale. This advantage comes from an ecosystem built over years and nearly impossible for competitors to replicate.

As the company's blog post puts it: "When enabled, Gemini accesses your data to answer your specific requests and to do things for you. And because this data already lives at Google securely, you don't have to send sensitive data elsewhere to start personalizing your experience. This is a key differentiator."

The app connections to Gemini are off by default; users must enable each app individually. According to Google, photos and emails aren't used directly for training but are referenced. Training relies on prompts and model responses after personal data has been "filtered or obfuscated."

Google has a new fancy name for when its AI gets things wrong

Google acknowledges the beta version is prone to bullshit. Users may run into inaccurate answers or what the company calls "over-personalization"—when the model draws connections between topics that have nothing to do with each other. Woodward gives an example: Hundreds of photos taken on a golf course could lead Gemini to assume the user loves golf. In reality, he's just tagging along with his son. The system doesn't pick up on that nuance.

The accompanying white paper lists other problems: Gemini sometimes confuses family members, ignores life changes like divorces, and forgets corrections. Tell the system you don't eat steak, and you might still get steakhouse recommendations a week later. Users have to fix mistakes manually or flag them with a "thumbs down."

OpenAI is chasing a similar vision with ChatGPT. The company wants to turn ChatGPT into an "intuitive AI super assistant" that doesn't just answer questions but actively takes on tasks and follows users across different channels. OpenAI CEO Sam Altman has said the goal is to build a personal assistant like the one in the movie "Her." For both companies, this also opens the door to more personalized advertising.