Type the word "disregard" into Google this week and the world's most-used search engine does not show you a dictionary entry. It thinks you're talking to it. On May 22, 2026, users searching for that single word watched Google's AI respond as though they'd issued a chatbot command.
This is what Google called its biggest upgrade in over 25 years. A search engine that can no longer be trusted to define a verb.
The disco-ball icons, the giant AI summary cards, the auto-generated paragraphs that materialize before the actual links — they are not a polish problem. They are a product problem. Google has been bolting a probabilistic word-predictor onto the front of a retrieval engine that used to be deterministic, and the seams are showing everywhere.
The errors stopped being funny a while ago
Early AI Overview gaffes had the texture of memes. Add glue to pizza. Drink two quarts of urine to pass a kidney stone. Easy to laugh at, easy for Google to patch.
Then came the medical stuff. The Guardian reported that AI Overviews told people with pancreatic cancer to avoid high-fat foods — the opposite of clinical guidance, and a recommendation experts said could increase the risk of patients dying from the disease. That is not a quirky hallucination. That is a search engine giving advice that could kill someone, served above the blue links a person was actually looking for.
And then the reputational damage. One property management company filed a complaint after Google's AI Overview attributed negative complaints from a different business to its verified listing, causing what the company called ongoing reputational and financial harm. Multiply that across every small business whose name happens to collide with a bad review in the training data.
The publishers are no longer asking nicely
Chegg has filed a US federal antitrust lawsuit arguing AI Overviews drastically reduce its traffic and revenue, and that Google is effectively coercing it into letting the AI consume its content without compensation. Penske Media Corporation — owner of Rolling Stone, Variety, and Billboard — has filed its own antitrust and content-use claim, alleging Google is using its journalism to power summaries that then keep readers from ever visiting the publishers.
In Europe, the Italian Federation of Newspaper Publishers has filed a complaint with Italy's communications regulator calling AI Overviews a "traffic killer" that harms media diversity and possibly breaches EU rules.
A pattern. The companies that produce the information Google's AI rewrites are now in court arguing the rewrite is the business model.
Google's defense is the problem
Liz Reid, who runs search at Google, has described the strategy as merging artificial intelligence and traditional web search to bring "the best of web and the best of AI together". In a separate blog post she acknowledged that "some odd, inaccurate or unhelpful AI Overviews certainly did show up".
Reread that. "Some." "Did show up." The framing treats hallucinated cancer advice and definition-of-a-word failures as edge cases in an otherwise healthy system. Reddit threads tell a different story: complaints about inaccurate summaries, inconsistent answers, and users unable to turn the feature off at all.
Google's own published guidance, meanwhile, still insists its ranking systems aim to reward original, high-quality content demonstrating expertise, experience, authoritativeness, and trustworthiness. The company is grading publishers on a rubric it is no longer applying to its own front page.
The real signal isn't the lawsuits or the blog posts. It's that people are leaving. Users are switching to Bing and DuckDuckGo specifically to escape the clutter. Google spent 25 years training the entire internet to type questions into one box. It is now teaching them, one urine-flavored kidney stone tip at a time, that the box has stopped working.
Search wasn't broken. Google broke it on purpose, because the AI race demanded a flag in the ground, and the flag had to go somewhere. It went on the homepage.
