Google's AI Overview boasts an accuracy rate of approximately 90%. Considering that Google handles over 5 trillion searches each year, this implies that the system could potentially generate around 57 million incorrect responses every hour. According to an evaluation conducted by the startup Oumi, Google's Gemini 2 model had an accuracy rate of 85% in October of the previous year. This figure improved to 91% for the Gemini 3 model in February of this year. However, it's important to note that Oumi's assessment methodology, which relies on AI tools, may introduce data bias. Consequently, the same search query might yield different overviews. Additionally, the proportion of AI-generated overviews that deviate from the original information increased from 37% for Gemini 2 to 56% for Gemini 3. These overviews are also susceptible to manipulation by misinformation, leading to inconsistencies and raising user concerns about their reliability. A Google spokesperson has questioned Oumi's testing methods, arguing that they do not accurately represent user search behavior.