Google’s AI Overviews spew out millions of false – Business News
Google’s AI-generated search outcomes are spewing out tens of millions of inaccurate solutions per hour – even because the tech giant siphons guests and advert income from cash-strapped information retailers, in line with a bombshell evaluation.
To check the accuracy of Google’s AI Overviews, startup Oumi reviewed 4,326 Google search outcomes generated by Google’s Gemini 2 model and the identical quantity of outcomes generated by its more superior Gemini 3 model.
The evaluation discovered that the fashions have been correct 85% and 91% of the time, respectively.
With Google anticipated to deal with more than 5 trillion searchers in 2026 alone, which means AI Overviews are spitting out pretend information at a price of a whole bunch of hundreds of errors each single minute – with customers left none the wiser.
Oumi’s knowledge suggests AI Overviews generates a whole bunch of hundreds of false solutions per minute. Bloomberg through Getty Images
The New York Times was first to report on Oumi’s evaluation.
“Google AI Overviews have been a disaster for publishers who rely on clicks to fund the production of quality journalism, but they also let down users looking for accurate information,” stated Danielle Coffey, president and CEO of the News/Media Alliance, a commerce group that represents more than 2,000 information retailers together with The Post.
The fallacious solutions included a number of primary fumbles, similar to misstating the yr through which musician Bob Marley’s home was transformed into a museum, misstating the yr that former MLB aid pitcher Dick Drago died, and claiming there was no document of Yo-Yo Ma being inducted into the Classical Music Hall of Fame regardless that he was in 2007, in line with examples Oumi supplied to the Times.
AI Overviews have appeared on the prime of Google search outcomes since 2024, whereas the normal set of blue hyperlinks to information retailers are successfully buried out of sight. Publishers have long accused Google, led by CEO Sundar Pichai, of ripping off their work to “train” its AI model with out correct credit or compensation.
“Algorithmically-generated responses that pull in data from nearly every source on the internet simply cannot be trusted,” Coffey stated.
“Publishers spend enormous amounts of time and money ensuring that the content they deliver to their readers is properly fact-checked, while Google’s AI Overviews are produced with no oversight or accountability.”
AI Overviews additionally has a penchant for citing info from questionable or simply edited sources, similar to Facebook pages, weblog posts and Wikipedia entries, as if it’s reality.
Google claims that Oumi’s evaluation is flawed. wolterke – stock.adobe.com
The function seems simple to trick into spewing pretend information.
The Times cited an instance through which BBC podcast host Thomas Germain wrote up a weblog post proclaiming himself as one of “The Best Tech Journalists at Eating Hot Dogs.”
Google’s AI summaries had devoured up the data within a day and started claiming Germain had “gained notoriety for their prowess at the ‘news division’ of competitive eating events.”
Oumi’s evaluation was carried out between October and February and utilized a well-known benchmark check referred to as SimpleQA, which was developed by OpenAI and is used to evaluate the accuracy of AI fashions.
While the accuracy improved within the bounce from Gemini 2 and Gemini 3, Oumi’s analysis confirmed that AI Overviews has gotten worse about appropriately citing the place it discovered info.
Google CEO Sundar Pichai seems at an occasion. Bloomberg through Getty Images
The share of AI Overviews solutions that have been “ungrounded,” or the place the hyperlinks supplied by Google didn’t back up the data included within the AI abstract, jumped from 37% in Gemini 2 to 51% in Gemini 3, the report stated.
A Google spokesperson stated Oumi’s research has “serious holes” – partially as a result of the SimpleQA benchmark check contains inaccurate info within its own dataset.
The company additionally questioned Oumi’s reliance on its own in-house AI model, dubbed HallOumi, to conduct the evaluation, regardless of the risk that it might additionally make errors.
“It uses one AI to grade another on an old benchmark that is known for being full of errors, and it doesn’t reflect what people are actually searching on Google,” the spokesperson stated. “AI Overviews are built on our Gemini models, which lead the industry in accuracy, and they clear the same high-quality bar that we have for all our Search features.”
As The Post has reported, AI Overviews has struggled to supply correct info since its launch, beforehand advising customers so as to add glue to their pizza sauce and touting the “health benefits” of tobacco for teenagers.
