When you search Google today, you're increasingly greeted by AI-generated summaries sitting atop your results. They look polished. They sound confident. But according to reporting from the New York Times, what's powering them isn't always what you'd expect.

Google's AI Overviews don't just pull from established news outlets or peer-reviewed research. They're also drawing on Facebook posts, Reddit threads, and other user-generated content — sources that carry no editorial oversight and often contain speculation, misinformation, or outdated information.

The problem isn't that Google uses diverse sources. It's that the AI presents all of this information with the same authoritative tone, regardless of origin. A medical claim sourced from a forum post gets the same visual treatment as one from the Mayo Clinic. Users have no easy way to distinguish between them.

This matters because Google's interface encourages trust. The AI Overview sits in a prominent blue box at the top of search results, a position traditionally reserved for the most relevant answers. It doesn't flag uncertainty or highlight when it's synthesizing contradictory sources.

Who's actually checking this?

Google says its systems are designed to prioritize quality, but the company hasn't disclosed how it weights different source types or what safeguards prevent low-quality content from shaping answers. As reported by the Times, the opacity around sourcing decisions leaves users — and even researchers — guessing about reliability.

The stakes are real. People use search for consequential decisions: health symptoms, financial advice, legal questions. When AI summaries flatten the difference between expert consensus and random internet commentary, the risk isn't just inaccuracy. It's misplaced confidence.

Google isn't alone in grappling with these issues. Every company racing to deploy generative AI faces similar tradeoffs between speed, scale, and trustworthiness. But Google's dominance in search gives it outsize influence over how billions of people access information.

The question isn't whether AI can summarize the web. It's whether it should — and whether users understand what they're actually getting when it does.