1 hour ago
A caller study recovered that these recaps regularly supply incorrect accusation and incorporate facts not supported by cited sources.
A German tribunal has ruled that Google is straight liable for incorrect accusation presented by its AI Overviews platform, according to a study by The Decoder. The state has laws successful spot that protect hunt motor operators from liability, but nan tribunal ruled that this doesn't use to AI overviews. It has classified Google arsenic a nonstop infringer because nan AI Overview is its ain contented and not conscionable a database of hunt results.
This each started erstwhile nan company's AI overview algorithm dispersed mendacious claims astir 2 Munich-based publishers. The publishers were tied to scams, subscription traps and shady business practices via definite hunt queries. The tribunal says nan AI jumbled up accusation astir wholly abstracted companies, drafting connections that didn't look successful immoderate linked sources and didn't really exist.
The publishers sent Google a cease-and-desist letter, but they opportunity it didn't respond appropriately. The Regional Court of Munich has deed Google pinch a impermanent injunction successful which it is nary longer allowed to dispersed mendacious accusation astir nan 2 companies progressive successful nan case.
The ruling places nan onus of work for immoderate actual errors connected Google, arsenic nan AI Overview rewrites accusation "in its ain words and according to its ain structure." In this case, nan overview confidently suggested that 1 of nan publishers was "known for dubious business practices" and built its ain building pinch a summary, reddish flags for these shady practices and tips for users. The problem, again, is that AI Overview was really pulling accusation astir different entity. It moreover invented claims retired of bladed aerial that weren't noted successful hunt results.
Simply put, Google nan hunt motor wouldn't beryllium straight liable because each it does is make third-party contented findable. Google nan AI Overview usability is liable because these overviews create "independent, caller and substantive statements" by evaluating and combining contented from third-party websites. This is beautiful overmuch what I do for work, and I surely would beryllium liable if I made up a bunch of slanderous worldly astir a mates of publishing companies.
Google based on astatine nan proceeding that users could cheque nan linked sources to verify if nan AI summary was correct. It besides said that these users knew "that accusation generated pinch AI should not beryllium blindly trusted." That's a reasonably singular connection fixed nan velocity successful which these Gemini overviews were foisted upon us. Studies propose that conscionable one percent of users click connected root links aft reference 1 of these overviews.
"Google users are little apt to click connected links erstwhile an AI summary appears successful nan results" according to a caller study by @pewresearch: https://t.co/eFGpniAP6m pic.twitter.com/812XtjqLN5
— Bob Pickard (@BobPickard) July 22, 2025
Just really bad is nan AI Overview actual accuracy problem? This depends connected your perspective. A caller study reported connected by The New York Times suggested that nan overview gets worldly incorrect astir 9 percent of nan time. That sounds for illustration a decent capable metric until you recognize conscionable really engaged nan level is. Google recently crowed that 2 cardinal group interact pinch AI Overviews each month. That's 24 cardinal group each year. Some elemental napkin mathematics suggests that translates to good complete 2 cardinal incorrect queries each year. This is an highly blimpish estimate, fixed really ubiquitous nan level has go and really galore Google searches (16.5 billion) are performed each day.
However, actual accuracy isn't nan only problem here. A study noted that these overviews besides person a sourcing issue. This study recovered that 56 percent of correct answers couldn't really beryllium backed up by nan linked source. In different words, there's nary existent measurement for users to really cheque nan AI's work.
As an aside, location is simply a measurement for users to "miss out" connected each of this magical AI hunt technology. Just popular "-ai" astatine nan opening of immoderate Google query. You're welcome.
