"Oumi's analysis showed that Google's AI Overview feature produces hundreds of millions of incorrect answers every day, with a 9% inaccuracy rate translating to approximately 225 billion false summaries annually."
"56% of the accurate answers produced by AI Overview were 'ungrounded,' meaning they cited sources that did not support their assertions, with a notable increase in ungrounded responses in Gemini 3."
"Of the 5,380 sources referenced by Google's AI summaries, Facebook and Reddit were among the most-cited, raising concerns about the reliability of the information provided."
"Google acknowledged that AI models can make mistakes but argued that Oumi's analysis was based on a benchmark test that contains flaws and inaccurate data."
An analysis by Oumi revealed that Google's AI Overview feature produces hundreds of millions of incorrect answers daily. The evaluation of 4,326 searches showed 85% accuracy with Gemini 2 and 91% with Gemini 3. Even a 9% inaccuracy rate translates to approximately 225 billion false summaries annually. Additionally, 56% of accurate answers were ungrounded, with a notable increase in ungrounded responses in Gemini 3. Sources like Facebook and Reddit were frequently cited, raising concerns about reliability. Google acknowledged the mistakes but criticized the analysis for using a flawed benchmark.
Read at TechRepublic
Unable to calculate read time
Collection
[
|
...
]