Google AI Overviews Wrong 1 in 10 Cases, New Analysis Finds
Google’s AI Overviews gave incorrect answers in about 10% of cases, according to a New York Times-commissioned analysis of 4,326 searches by Oumi, despite switching to the Gemini 3 model. Accuracy rose to 91% in February, but over half of correct answers cited sources that did not fully support the claims. Google disputed the findings, calling the study flawed. Pew found users clicked traditional search results less often when AI summaries appeared.