Particle.news

Google’s New AI Overview Mistakes Command Words for Instructions

The Gemini‑powered AI layer misclassified single‑word queries as chat instructions, and Google says it is rolling out a fix after public reports of the problem.

Overview

  • Users and reporters noticed on May 22 that searches for single verbs like “disregard,” “ignore,” “dismiss,” “stop,” “start,” “skip,” and “quit” produced chatbot‑style replies in the AI Overview box instead of showing a dictionary definition or the usual top links.
  • The AI Overview response often consisted of a brief conversational reply followed by a large blank space that pushed standard blue‑link results, including Merriam‑Webster definitions, below the fold and out of immediate view.
  • The bug appeared unevenly across browsers and accounts, with some people seeing the chatbot reply in Safari or on mobile while others using Chrome or different accounts saw normal results, consistent with a staged rollout or A/B test.
  • Google acknowledged the misinterpretation, told multiple outlets the problem is unrelated to the I/O changes, and said engineers are working on a fix that will roll out soon; at least one outlet reported the AI Overview was temporarily paused during remediation.
  • The incident highlights a recurring risk of putting generative AI at the top of search: it can change basic user flows, reduce immediate referrals to source sites, and raise fresh concerns about reliability and web traffic for publishers as Google expands Gemini‑powered Overviews.