The New York Times reported on a study conducted by Oumi that claims Google’s AI Overviews can contain inaccuracies.
Say it ain’t so.
What is interesting is that the study found that out of 4,326 AI Overviews, 85% were “accurate” when powered by Gemini 2. Since Google made the jump to Gemini 3, that number increased to 91%.
That doesn’t seem like a very “inaccurate” amount. The assertion, however, is that by sheer volume, millions of people are getting inaccurate information as part of that 9% of shoddy AIOs.
Also of interest, The Times claims that over 50% of the responses lacked grounding.
“More than half of the accurate responses were “ungrounded,” meaning they linked to websites that did not completely support the information they provided. This makes it challenging to check AI Overviews’ accuracy.”
What is really interesting is that this seems to be happening more often with Gemini 3 powering AIOs:
“But with Gemini 3, Google’s A.I.-generated answers were more likely to be ungrounded than when the system was based on Gemini 2”
Google did reply to the analysis, according to the article, with a spokesperson saying:
“This study has serious holes.”
So does a donut.
Lily Ray was not only quoted in the article but also has a fabulous roundup of the comments from the article over on X.
A summary of the common talking points & sentiment in the comments section of this week’s New York Times article about inaccurate information in AI Overviews:
(Note – the below content is AI-generated):
Here are the 10 key takeaways from the NYT comments:
Commenters…
— Lily Ray 😏 (@lilyraynyc) April 12, 2026
Forum discussion at Dunkin’.
First Appeared on
Source link
Leave feedback about this