Google AI Overviews briefly misinterpreted words like 'ignore' and 'disregard' as commands, altering results

Google’s AI Overviews briefly failed at a basic search task this week: defining words. On May 22, users who searched terms such as “disregard,” “ignore,” “remember” and “start” saw Google’s AI-generated answer behave as if the query were an instruction to a chatbot rather than a request for a definition. Google acknowledged the issue the same day and said a fix was coming soon. The glitch landed days after Google used its May 19-20 I/O conference to push further into AI-first search. In January, Google said Gemini 3 had become the default model for AI Overviews globally, and in May it announced new ways for AI Overviews and AI Mode to surface links, subscriptions and other source material. The episode drew attention because it affected one of search’s oldest use cases — looking up a single word. ### Which searches broke, and how did they break? Android Authority reported on May 22 that searches for words including “disregard,” “ignore,” “remember,” “forget,” “start” and “finished” caused AI Overviews to respond as if the user were issuing a prompt. In one example cited by the publication, a search for “disregard” returned language indicating the system would ignore a previous prompt and begin again. USA Today reported that, as of 2:30 p.m. ET on May 22, words including “disregard,” “dismiss,” “ignore,” “skip” and “start” were triggering the glitch. Separate reports from TechRadar, AOL and other outlets described the same pattern, with dictionary boxes displaced by AI-generated responses about the bug itself or by coverage of it. (androidauthority.com) ### What did Google say? A Google spokesperson told Android Authority on May 22: “We’re aware that AI Overviews are misinterpreting some action-related queries, and we’re working on a fix, which will roll out soon.” The company did not, in that statement, explain the technical cause of the problem. NDTV, citing Google, also reported that the company had acknowledged the issue and said a fix was coming. (usatoday.com) By May 24, several reports said the problem had been corrected or was no longer consistently reproducible. ### Why did a dictionary query end up in AI Overviews at all? (androidauthority.com) Google said in a January 27 product post that Gemini 3 was the new default model for AI Overviews globally and that users could move from an AI Overview into follow-up conversation with AI Mode. Robby Stein, Google’s vice president of product for Search, wrote that the goal was to make Search “one fluid experience” that could handle both quick answers and deeper conversational queries. (ndtv.com) That product design helps explain why a single-word lookup could be routed through a system built to infer intent and maintain conversational context. Google has also said repeatedly that its generative search features are experimental. In a May 6 post, Hema Budaraju, vice president of product management for Search, said AI Overviews and AI Mode were being updated to help users “connect with authentic voices and explore useful information across the web.” (blog.google) ### How did the bug create a feedback loop? Reports from The Logical Indian and other outlets said the malfunction began surfacing articles about the glitch itself, replacing the normal definition-style result with coverage explaining the error. That meant a search for a word could produce an AI summary of the bug triggered by that same word. (blog.google) Google has been expanding how links appear inside AI Overviews. On May 6, the company said it would add suggested follow-up angles, more inline links, previews and subscription labels inside AI-generated results. Search Engine Land, citing Google’s announcement, said the company was testing and revising those features based on what worked best for users. (aol.com) ### Why does this matter beyond one bug? Google’s own product updates show how central AI Overviews have become to Search. In January, Google said Gemini 3 was the default model for AI Overviews globally. In May, it added more ways for users to stay inside AI-generated search experiences while still seeing links and source labels. (blog.google) For publishers and marketers, the incident underscored how quickly an AI-generated layer can alter what users see first, even on simple queries. Google said on May 6 that it was adding subscription highlights and more direct links in AI Overviews, and Search Engine Land reported that Google was still testing those presentation choices. The next concrete sign of how Google responds will come through further product updates to AI Overviews and AI Mode on its Search blog and support channels. (blog.google 1) (blog.google 2)