Google AI Overview Accuracy at Approximately 90% Produces Tens of Millions of Incorrect Answers Per Hour
Oumi's analysis shows that the accuracy of Google's "AI Overview" feature is about 90%. Given that Google processes over 500 trillion searches annually, this means that tens of millions of incorrect answers are generated every hour (hundreds of thousands of inaccurate pieces of information per minute).
The accuracy with Gemini 2 was 85%, while Gemini 3 has improved to 91%. Among the 5,380 sources analyzed, Facebook and Reddit were the second and fourth most cited sources, respectively, with a higher proportion of inaccuracies attributed to Facebook (7% vs. 5% when accurate).
Source: Public Information
ABAB AI Insight
Google's AI Overview has rapidly iterated the Gemini model since its launch in 2024. This Oumi analysis, based on the SimpleQA benchmark, continues the transition from experimental AI search to core product. It has previously faced user and regulatory criticism due to several high-profile incorrect answers (such as fabricated information or outdated data).
In terms of capital strategy, Google increases search dwell time and ad displays through AI Overview, but the massive output of errors at a 90% accuracy rate directly amplifies trust costs. The high citation rates from UGC sources like Facebook and Reddit further expose weaknesses in the fact-checking mechanisms. Strategically, Google needs to continuously invest resources to optimize source weighting and verification layers.
In comparison to tools like Perplexity, Claude search, or traditional Google blue links, the AI-enhanced search market is transitioning from speed-first to reliability and source transparency. Tools with strong fact-checking capabilities are gaining pricing power.
Essentially, this represents a technological substitution: AI Overview shifts search from manual curation to automated generation, but hallucinations and low-quality UGC citations shift trust pricing power from the platform to user judgment. The mechanism lies in the scaled output amplifying the impact of a single error, accelerating capital concentration in AI search entities that can effectively control accuracy, source quality, and transparency.
ABAB News · Cognitive Law
The larger the search volume, the more a 90% accuracy rate translates to hundreds of thousands of errors per minute; scale is always an amplifier of defects.
The more UGC sources are cited, the easier it is for AI Overview to be polluted by social noise; transparent sources are the true barrier.
The faster the model iteration, the more covert the loss of user trust; accuracy is always the ultimate lever for retention.