A new analysis by The New York Times has raised concerns about Google’s AI Overviews feature. The report claims that the system can generate millions of errors per hour, even with a very high accuracy rate.
AI Overviews by Google were introduced in 2024 to provide direct answers on the Google Search page.
The latest study found AI Overviews are accurate about 90% of the time. One in ten responses still contains false information. Google handles nearly 5 trillion searches each year. That scale results in over 57 million incorrect answers each hour, or nearly 1 million every minute.
Researchers said the issue is not accuracy alone. The real concern is how quickly errors spread at such a scale.
The study used testing by startup Oumi with the SimpleQA benchmark. It evaluates factual accuracy in generative AI systems like Gemini.
The results show progress. AI Overviews recorded about 85% accuracy with Gemini 2.5 last year. The rate improved to 91% after the Gemini 3 update.
Researchers still flagged risks across categories. Errors appeared in factual queries, service details, and even health-related topics. Experts warned that such mistakes can mislead users and affect decision-making.
Also Read: Google Gemini Adds Mental Health Crisis Support to AI Assistant
Google has pushed back against the findings. The company said the study has gaps and depends on a benchmark built by OpenAI that may include incorrect data.
“This study has serious holes,” a Google spokesperson said.
AI Overviews have faced criticism earlier as well. During the Air India crash in Ahmedabad, the feature misidentified the aircraft model. The error triggered backlash online. Google later removed the response and said it uses such cases to improve its systems.
Although the rollout reflects a shift in search, the debate now centers on balancing speed with accuracy.