Google's AI Overviews Feature Generates Millions of Incorrect Responses Every Hour

In 2024, Google introduced a groundbreaking feature known as 'AI Overviews' that prominently displays answers generated by artificial intelligence at the top of the search results page. This innovation has transformed Google's role from merely curating information to acting as a publisher, prompting significant concerns regarding the accuracy of the data being presented.

According to a recent analysis conducted by the startup Oumi in collaboration with The New York Times, the accuracy of the responses provided by the AI Overviews feature stands at approximately nine out of ten cases. However, considering that Google processes over five trillion search queries annually, this translates to millions of incorrect responses being generated every hour, equating to hundreds of thousands of inaccuracies every minute.

The study revealed that more than half of the accurate responses were deemed 'unsubstantiated,' as they referenced websites that did not fully corroborate the information provided. This raises red flags about the reliability of the data users receive through this feature.

At the request of The New York Times, Oumi conducted an accuracy analysis of the AI Overviews responses using a test called SimpleQA, which is widely accepted in the industry for measuring the accuracy of AI-based systems. The startup tested Google's system in October, when the most complex queries were addressed using an AI called Gemini 2, and then repeated the test in February after an upgrade to the more powerful Gemini 3 technology.

In both instances, Oumi focused on 4,326 Google search queries. The verification results indicated that the accuracy of the responses was 85% for Gemini 2 and 91% for Gemini 3. This suggests some progress in the development of the technology; however, a significant portion of errors remains.

Google acknowledges that the AI Overviews feature may contain inaccuracies. In fine print beneath each AI overview, it states: 'Artificial intelligence can make mistakes, so please verify the answers again.' This reminder underscores the necessity for a critical approach to the information obtained.

However, Google representatives contended that Oumi's analysis was flawed, as it relied on a benchmark test created by OpenAI, which itself contained incorrect information. This has sparked discussions about the reliability of the methods used to assess the accuracy of artificial intelligence.

Pratik Verma, CEO of Okahu, a company specializing in helping individuals understand and utilize artificial intelligence technologies, believes that Google's accuracy is comparable to that of any leading AI systems. He also urged users to verify the information received to avoid errors.

It is worth noting that earlier, American company Google launched a new AI-based feature for electric vehicles in Google Maps. This move signifies the ongoing evolution of artificial intelligence technologies across various sectors; however, it is crucial to remain aware of the risks associated with their use.

Join our channels on Telegram, Instagram, and YouTube to stay updated on the latest news and events in the world of technology.