Google AI Search Summaries Reach 91% Accuracy but Could Still Produce Tens of Millions of Wrong Answers Daily

Google has integrated its Gemini generative AI into Search, using AI Overviews to synthesize information from webpages and answer questions directly. Given Google's vast daily search volume, even a small error rate could be amplified at scale despite accuracy approaching 90%, potentially affecting users' assessment of historical information, news and fact-checking material.

The latest testing showed that Google AI Overviews answered 91% of questions correctly. Applied across total search volume, however, the system could still produce tens of millions of incorrect answers each day. Test cases found that it continued to make mistakes when identifying historical dates and verifying objective facts. The reports did not disclose the test date or full sample size, leaving the findings in need of further validation.