Investigation Finds Google Search's 'AI Overviews' Produce Tens of Millions of False Summaries Per Hour

The New York Times and Oumi tested the accuracy of Google Search's AI Overviews using the SimpleQA benchmark.

When the feature ran on Gemini 2.5 the accuracy was 85%, and it rose to 91% after upgrading to Gemini 3.

NYT noted Google handles over 5 trillion searches yearly and estimated a 90% accuracy rate would produce tens of millions of false summaries per hour.

Google said SimpleQA contains incorrect data and does not reflect real user search behavior.

NYT warned that web content can be manipulated to make AI Overviews present misleading expert claims.