Testing suggests Google's AI Overviews tells lies per hour (arstechnica.com) AI

A test analysis (via Oumi) that benchmarks Google’s AI Overviews against thousands of fact-checkable questions found it answers correctly about 90% of the time, implying large numbers of incorrect summaries across all searches. Examples cited include confident factual errors about dates and institutions. Google disputes the benchmark’s relevance, saying the test includes problematic questions and that it uses different models per query to improve accuracy.