Study Reveals Google’s AI Overviews Spreads Significant Misinformation

Google’s AI Overviews Feature Produces Millions of Errors Hourly, Study Finds

A recent analysis has revealed that Google’s AI Overviews feature, which provides summary information above search results, delivers incorrect information at an alarming rate despite maintaining a seemingly impressive accuracy percentage.

The study, conducted by AI startup Oumi for The New York Times, found that Google’s AI Overviews were correct approximately 90 percent of the time. While this figure appears strong at first glance, the sheer volume of Google’s search traffic—roughly five trillion queries annually—means that even this error rate translates to tens of millions of incorrect answers every hour and hundreds of thousands every minute.

Even more concerning, researchers discovered that more than half of the technically accurate responses were “ungrounded,” meaning they linked to websites that didn’t fully support the information provided in the overviews. This disconnect makes it difficult for users to verify the accuracy of the AI-generated summaries, potentially contributing to what both The New York Times and technology news site Futurism have described as a “misinformation crisis.”

To evaluate the accuracy of the AI Overviews feature, Oumi employed a benchmark test called Simple QA, which is widely used in the industry to measure AI system accuracy. The analysis tested two versions of Google’s AI system: Gemini 2 in October and the upgraded Gemini 3 in February. Across 4,326 searches, Gemini 2 produced correct answers 85 percent of the time, while the newer Gemini 3 showed improvement with a 95 percent accuracy rate.

Google has acknowledged these issues but has attempted to minimize their significance. “Our Search AI features are built on the same ranking and safety protections that block the overwhelming majority of spam from appearing in our results,” said Google spokesperson Ned Adriance in a statement to The New York Times. Adriance further claimed that the benchmark test “doesn’t reflect what people are actually searching on Google.”

The findings come at a critical time for Google, which has been aggressively expanding its AI capabilities to maintain its dominance in the search market while competing with emerging AI-powered alternatives. The company recently expanded its AI Overview feature to the Canadian market, part of its global rollout strategy.

The revelation raises serious questions about the responsibility of tech giants in deploying AI systems at scale. With billions of users worldwide relying on Google for information, even a small percentage of errors can have significant real-world impacts across education, health information, financial decisions, and other critical domains.

AI experts have long warned about the challenge of hallucinations—instances where AI systems confidently generate false information—in large language models like those powering Google’s AI Overviews. This study provides concrete evidence of how these theoretical concerns manifest in widely used consumer products.

For users, the findings underscore the importance of critical thinking when consuming AI-generated content and the continued need to verify information through multiple sources, particularly for consequential decisions or sensitive topics.

As AI becomes increasingly integrated into our information ecosystem, this study highlights the tension between rapid technological advancement and the need for accuracy and reliability in systems that millions depend on daily for access to knowledge.