New York Times analysis finds 10% of Google AI answers wrong
Updated
Updated · The New York Times · May 7
New York Times analysis finds 10% of Google AI answers wrong
10 articles · Updated · The New York Times · May 7
The report said errors included basic facts such as historical dates, names and celebrity news, implying tens of millions of mistaken answers each hour across Google's search volume.
The article said Google's AI search can still speed tasks like identifying car parts, choosing condiments and spotting scams when users provide limited information rather than broad web queries.
Google disputed the study, saying most AI responses are accurate and the analysis did not reflect typical searches, while acknowledging the technology can misinterpret web content.
Google’s AI is fast but frequently wrong. Is this popular new tool ultimately more helpful or harmful?
With Google's AI eliminating website clicks, how can the internet's original content creators possibly survive?
The Hidden Crisis in Google AI Overviews: Tens of Millions of Daily Errors Despite 90% Accuracy
Overview
A 2026 study revealed that Google's AI Overviews, powered by Gemini 2 and 3 models, achieve 85-91% accuracy but still produce hundreds of millions of errors daily due to the massive volume of over 5 trillion annual searches. These errors include false information and harmful advice, often stemming from flawed source selection and limitations of the AI models. The AI frequently cites unreliable sources, which undermines user trust and reduces engagement with original content. Despite Google's efforts to improve models and contest findings, experts remain concerned about the scale of errors and lack of transparency. Experts recommend systemic solutions and urge users to critically verify AI-generated summaries to mitigate risks.