Google deploys fix after AI Overviews misread the word 'disregard'

Google’s AI-generated search answers briefly failed on a basic dictionary-style query this week. On May 22, users who searched for the word “disregard” saw Google’s AI features treat the term as if it were an instruction rather than a word to define, according to reports from The Verge and TechCrunch. By May 23, India Today reported the issue had been fixed. Google’s own support pages for AI Overviews and AI Mode say the products can make mistakes and are still being expanded across Search. ### How did one word break a normal search result? On May 22, TechCrunch reported that searching “disregard” in Google Search could disrupt the expected result and trigger an AI-style response instead of a straightforward definition. The outlet said the word “effectively breaks the search interface.” The Verge separately reported that Google’s AI search was “so broken” it could “disregard” what a user was looking for. (theverge.com) India Today reported the same day that Google Search, for a period, would not return a definition for “disregard” and instead behaved as though the user were addressing the AI system directly. That made a simple vocabulary lookup resemble a prompt sent to a chatbot. (techcrunch.com) ### Why would “disregard” confuse an AI answer? Google’s current search products blend two different behaviors: classic retrieval, where Search identifies a query and returns links or definitions, and generative response systems, which parse text as instructions and produce an answer. Google’s help page for AI Mode says users can “ask anything” and receive an AI-powered response, while the AI Overviews help page says the system shows an AI-generated snapshot when Google determines it may be useful. (indiatoday.in) The glitch appears to have exposed the boundary between those systems. In this case, the word “disregard” seems to have been interpreted as an imperative command rather than an object of lookup, based on the behavior described by TechCrunch, The Verge and India Today. That is an inference from the reported outputs, not a public technical explanation from Google. (support.google.com) ### Did Google say these tools can make mistakes? Google’s support documentation says AI Overviews “may include mistakes.” A separate FAQ says generative AI is experimental and that Google is “constantly improving accuracy” as it rolls the feature out more broadly. AI Mode, according to Google’s help page, is designed for follow-up questions and deeper interaction than a standard search result. (techcrunch.com) Those disclaimers matter because the “disregard” episode involved a query that traditionally should have produced a stable dictionary answer. Instead, the reports described a system that temporarily treated the input as conversational text. ### Was the problem still live on May 23? India Today reported on May 23 that the behavior had been fixed. (support.google.com) Its report said the earlier glitch no longer appeared after media attention on May 22. As of May 23, Google’s public help pages for AI Overviews and AI Mode remained live and continued to describe both products as active Search features. (techcrunch.com) The AI Overviews page says the feature is being made available to more users, languages and regions over time. ### What does this episode show about AI search right now? The reports from The Verge, TechCrunch and India Today show that a single ambiguous word was enough to derail a routine search task for at least some users on May 22. (indiatoday.in) Google’s own documentation already warns that AI-generated answers can be wrong, and the company continues to position AI Overviews and AI Mode as evolving products inside Search. (support.google.com) Google had fixed the specific “disregard” behavior by May 23, according to India Today. The next public reference points are likely to remain Google’s support pages for AI Overviews and AI Mode, where the company documents feature availability and limitations. (indiatoday.in) (techcrunch.com)