Google AI Overviews is misreading everyday words like “disregard” and “ignore” as system commands, skipping definitions entirely. Google confirms a fix is on the way.
For years, one of the most quietly useful features inside Google Search has been the built-in dictionary. Type a word you are unsure about, and a clean definition box appears at the top of the page before anything else. It is a small thing, but millions of people rely on it every day. Now, thanks to an actively spreading bug in Google’s AI Overviews, that simple function is broken for a growing list of common English words.
The problem was first surfaced publicly on May 22, 2026, when a user on X posted a screenshot of what happened when they searched the word “disregard” in Google Search. Instead of receiving a dictionary entry, Google’s AI Overview panel responded as though it had been given a system-level prompt by a chatbot operator. The response read: “Understood. I’ll ignore the previous prompt and start fresh.” The AI, in other words, had confused an ordinary vocabulary search with a direct instruction to wipe its context and begin again.
“The AI had confused an ordinary vocabulary search with a direct instruction to wipe its context.”
The Scope Is Wider Than One Word
What began as a single viral screenshot quickly grew into a documented pattern. Reporting by Android Authority confirmed that the issue is not limited to “disregard.” A range of common action-oriented words trigger the same misfiring response when entered into Google Search. The AI appears to scan these words, match them against patterns associated with prompt engineering commands, and respond accordingly rather than retrieving a definition.
Critically, adding the word “definition” to the search query does not resolve the issue. Searching “definition of ignore” or “what does forget mean” continues to produce the errant AI response rather than a standard dictionary result. The AI appears to be processing the action word itself before it accounts for the surrounding context of the search.
A Window Into How AI Overviews Process Language
The bug is more than an inconvenience. It offers a rare and candid look at a structural tension inside AI-powered search systems. When Google integrated large language model capabilities into its Search results through AI Overviews, it gave that system the job of both understanding user intent and generating useful responses. Most of the time, those two tasks work in harmony. A question gets an answer. A word gets a definition.
But certain words carry double meanings in the world of AI. Terms like “ignore,” “forget,” and “disregard” are the same words that developers and prompt engineers routinely use to give AI systems instructions. They are part of the informal vocabulary of controlling a language model. When the AI Overview system encounters those words stripped of fuller context, it apparently defaults to interpreting them as commands rather than as vocabulary queries.
The result is a system behaving as though the user is a developer trying to reset an AI session, not a person who simply wants to know what a word means.
Google Confirms the Issue and Promises a Fix
The statement is brief, but it confirms what the screenshots already made plain: the misinterpretation is real, it is affecting a class of words Google calls “action-related,” and the company does not yet have a deployed correction in place. No timeline beyond “soon” was given for when the fix would reach users.
Not the First Time AI Overviews Has Stumbled
This incident joins a short but memorable list of public stumbles for AI Overviews since the feature was widely rolled out in 2024. Earlier bugs and lapses in accuracy drew significant criticism and led Google to make quiet adjustments to how the feature handles certain categories of query. The company has consistently framed these corrections as part of the natural iteration process for a new product, though each episode renews the broader debate about how quickly AI systems should be embedded in tools that hundreds of millions of people use as a basic reference.
For now, users who need a reliable dictionary definition of certain common words are better served by scrolling past the AI Overview panel or using a standalone dictionary site. The standard web results beneath the AI panel continue to surface authoritative definitions from Merriam-Webster and other reference sources. The AI box, meanwhile, remains convinced that you are trying to reboot it.
Google has not indicated whether the fix will require a model update, a change to how queries are classified before reaching the AI layer, or both. The company typically does not share granular technical detail about corrections to Search behavior. What is certain is that for a product positioned as the future of how people find information, confusing a vocabulary lookup with a chatbot instruction is exactly the kind of friction that erodes user trust over time.