Google’s AI Overviews Spreading Misinformation at Unprecedented Scale, Analysis Finds

Google’s AI Overviews Spreading Misinformation at Unprecedented Scale

Google’s AI-generated search summaries are delivering false information to millions of users every hour, according to a new analysis that raises serious concerns about the technology’s widespread deployment.

A recent study conducted by AI startup Oumi for The New York Times found that Google’s AI Overviews—the AI-generated summaries that appear at the top of search results—are accurate approximately 91 percent of the time. While this figure might appear impressive in isolation, the scale of Google’s search operation transforms it into a significant problem.

Google processes roughly five trillion search queries annually. This volume means that even with a 91 percent accuracy rate, the system is generating tens of millions of incorrect answers every hour—or hundreds of thousands every minute, according to Oumi’s calculations.

The findings are particularly troubling given research on how users interact with AI systems. Studies have consistently shown that people tend to accept AI-generated information without verification. One report found that only 8 percent of users actually double-check an AI’s answers, while another experiment revealed that users followed incorrect AI advice nearly 80 percent of the time—a phenomenon researchers have labeled “cognitive surrender.”

“Large language models adopt an authoritative tone and can confidently present fabricated information as fact,” the analysis noted. Combined with the convenience of Google’s AI Overviews, this creates a perfect storm for widespread misinformation.

Oumi conducted its analysis using SimpleQA, an industry benchmark for AI accuracy developed by OpenAI. Researchers tested the feature twice—first in October using Google’s Gemini 2 model, and again in February after Google upgraded to Gemini 3.

Each testing round involved 4,326 Google searches. The results showed improvement between iterations, with Gemini 3 achieving 91 percent accuracy compared to Gemini 2’s 85 percent. While this indicates progress, it also reveals that Google rolled out earlier versions of the technology knowing they were more prone to errors.

Google disputed the findings, with spokesman Ned Adriance telling the Times, “This study has serious holes. It doesn’t reflect what people are actually searching on Google.” However, the Times reported that Google’s own internal analysis found Gemini 3 produced incorrect information 28 percent of the time, though the company claims AI Overviews perform better because they incorporate Google search results before generating answers.

Perhaps most concerning is the issue of “ungrounded” responses—AI-generated answers that cite websites that don’t actually support the information provided. The analysis found this problem has worsened, with ungrounded responses increasing from 37 percent with Gemini 2 to 56 percent with Gemini 3. This trend not only suggests the AI is fabricating information but also makes it significantly harder for users to verify claims.

The findings come amid growing scrutiny of AI deployment by major tech companies. Microsoft recently faced criticism for terms of service language that described its Copilot AI as being for “entertainment purposes only,” seemingly undermining claims about the technology’s reliability for serious applications.

As Google and other tech giants continue embedding AI into products used by billions of people worldwide, the Oumi analysis highlights the tension between rapid AI deployment and ensuring information integrity. With only a small percentage of users verifying AI-generated content, even modest error rates can contribute to a substantial volume of misinformation circulating online.

The analysis raises important questions about responsibility and transparency as AI becomes increasingly integrated into how people access information online, potentially reshaping public understanding of facts on a scale unprecedented in human history.