Select Language:
The New York Times reported on a study by Oumi highlighting that Google’s AI Overviews may sometimes present inaccuracies.
It’s surprising, but the study found that out of 4,326 AI Overviews, 85% were accurate when powered by Gemini 2. After upgrading to Gemini 3, this accuracy rate rose to 91%.
While these percentages seem high, the core concern is that millions of users may be receiving incorrect information through that 9%. Additionally, over half of the responses examined lacked sufficient grounding. That is, more than 50% of the so-called accurate responses referenced sources that didn’t fully support the information provided, making it difficult to verify the truthfulness of AI-generated summaries.
Interestingly, this issue appears to be more prevalent with Gemini 3. The study noted that AI answers generated by Gemini 3 were more often ungrounded compared to those based on Gemini 2.
Google responded to the findings, dismissing the study as having “serious holes,” implying flaws in the analysis.
Lily Ray was quoted in the report and shared a comprehensive overview of public reactions on X (formerly Twitter), where users discussed the implications of AI inaccuracies.
A summarized snapshot of the comments suggested that many people remain concerned about the reliability of AI overviews, especially as new versions evolve and produce more ungrounded responses.
Forum discussions are ongoing, with many pondering the impact of these inaccuracies and what it means for everyday information consumption.
ChatGPT
Perplexity
Gemini AI
Grok AI
