Study: Google’s AI Overviews show millions of wrong answers every hour
Aside from hallucinations, immense energy requirements, and potentially negative mental health effects, generative artificial intelligence still has an accuracy problem. However, that hasn’t stopped big tech companies like Google from rolling out features like AI search summaries to their users. Most results look good at first glance and usually have multiple source citations, but that doesn’t mean the product works perfectly.
A recent study reported by the New York Times found that Google’s “AI Preview” delivers summaries that are correct and from trusted sources 9 times out of 10. But even though 90% seems like a passing grade, the failure rate adds up in minutes. Since the company will process more than five trillion searches in 2026, the resulting calculations mean that AI Overview produces tens of millions of questionable answers every hour. That’s hundreds of thousands of errors every minute.
Additionally, it is difficult to truly assess the accuracy of AI Overview. An initially incorrect answer to a search query can turn into a correct answer when a user repeats the search a second time, making it virtually impossible to predict. Google’s decision to place the AI ​​Overview tab at the top of most search results pages also means more people see it and can assume its trustworthiness.
Part of the problem is which specific sites are treated as sources. Authors of the study from open source AI company Oumi found that Facebook and Reddit were the second and fourth most cited references for AI Overview. Correct answers cited Facebook 5% of the time, while inaccurate responses cited the social media site 7% of the time. In other cases, AI Overview appears to misindicate a reliable source by giving the wrong answer.
Then there’s the problem of bad actors. Anyone who understands these bugs in the system can potentially cause AI Overview to give inaccurate statements. Hypothetically, a person could write a series of blog posts claiming downright false historical information and then artificially increase traffic to their website. Google’s AI preview may include the site in its source review, fail to flag its inaccuracies, and then generate the wrong response.
“Our Search AI features rely on the same ranking and security protections that prevent the overwhelming majority of spam from appearing in our results,” said Google spokesperson Ned Adriance. the New York Times. “Most of these examples are unrealistic searches that people wouldn’t actually do.”
At the very least, it’s important to view tools like AI Overview with a heavy dose of skepticism about the possible future. One wrong answer out of 10 answers might not seem too bad, but think about how many search queries you’ve already done today. Studies also already indicate that over-reliance on this type of technology may not be beneficial for overall cognitive abilities. But if there’s one thing you can trust, it’s AI Overview’s fine print disclaimer:
“The AI ​​can make mistakes, so check the answers.”
The best new PopSci 2025 releases
