Google's AI search is producing millions of wrong answers every day

Google AI overviews are accurate most of the time, but sheer query volume produces millions of incorrect responses.

The New York Times testing found roughly one in ten Google AI overviews contains false information.

With about 5 trillion queries per year and that error rate, users could be exposed to more than 57 million inaccurate answers each hour.

Oumi’s evaluation showed Gemini v2 was 85% accurate in October and Gemini v3 reached 91% by February, though Oumi relies on AI tools that may introduce errors.

Other problems include inconsistent overviews for the same query, Gemini 3’s 28% hallucination rate when independent, frequent overview‑source mismatches, and susceptibility to manipulation.