Google’s AI Overviews Are Making Mistakes at Massive Scale. Here’s What to Know
Google’s AI Overviews Are Making Mistakes at Massive Scale. Here’s What to Know
A new analysis shows millions of mistakes, creating risks for health, safety, and everyday searches.
BY LEILA SHERIDAN, NEWS WRITER
Illustration: Inc; Photo: Google
Starting in 2024, Google began putting AI-generated summaries—called AI Overviews—at the top of its search results. Instead of just pointing users to websites, Google started answering questions directly.
At Google’s scale, even small mistakes can quickly turn into much bigger problems, and as its AI rolled out, those errors started showing up almost immediately.
Study Claims 90% Accuracy
A recent analysis by AI startup Oumi found that AI Overviews were correct about 90 percent of the time. That sounds strong, until you consider that Google processes more than five trillion searches a year. At that volume, even a 10 percent error rate could translate into tens of millions of wrong answers every hour, The New York Times reported.
In some cases, those errors resulted in real-world consequences.
How Anthropic's Claude AI Became a Co-Founder
An investigation by The Guardian found that AI Overviews gave misleading information about liver blood test results. When users searched for “normal” ranges, they were shown lists of numbers without important context, like age, sex, or medical history. That could lead someone with serious liver disease to think their results are fine and skip follow-up care, The Guardian noted.
After the reporting, Google removed AI Overviews for those specific searches. A Google spokesperson told The Guardian that while the company doesn’t make changes based on individual cases, it works to improve the system more broadly when issues come up.
Scammers Exploit Problems
The risks aren’t limited to health information. According to Wired, scammers are exploiting AI Overviews. In some cases, users search for a company’s phone number and end up calling a fake number surfaced by the AI. The person on the other end then pretends to be the business and tries to collect payment or personal information.
