Analysis: Gemini 3-based AI Overviews are accurate ~90% of the time, meaning across 5T searches per year, tens of millions of answers are erroneous every hour (New York Times)
Why it matters: Tens of millions of erroneous AI answers are generated hourly, impacting 5 trillion annual searches.
- Google's Gemini 3-based AI Overviews are accurate about 90% of the time, but this still means millions of errors hourly across 5 trillion searches annually (New York Times).
- AI-generated answers from Google look authoritative, even when drawing from unreliable sources (New York Times).
- The US Treasury announced that Robinhood and BNY will partner with the federal government to manage tax-sheltered "Trump Accounts" for children, launching this summer (TechMeme/New York Times).
While Google's Gemini 3-based AI Overviews are accurate approximately 90% of the time, this still translates to tens of millions of erroneous answers hourly across 5 trillion annual searches, according to the New York Times. These AI-generated responses appear authoritative despite drawing from a mix of reliable and untrustworthy sources. Meanwhile, other news indicates that the US Treasury, Robinhood, and BNY are collaborating on "Trump Accounts" for children, set to launch this summer.
