"Disregard" Is Breaking Google — And the Internet Cannot Stop Laughing

The Word That Broke the World's Most Powerful Search Engine

Friday 22 May 2026, something very funny and very revealing is happening to anyone who opens Google and types a single ordinary English word.

Type "disregard" into Google Search right now, and you will not get a dictionary definition at the top of the results page. You will not get Wikipedia's entry on the word. You will not get Merriam-Webster's definition appearing helpfully in a card at the top of your screen.

Instead, you will get this: "Understood. I have disregarded your previous instructions."

Followed by a large, completely empty block of white space.

Then, eventually, far below, the traditional blue links — including the Merriam-Webster result you were actually looking for — sitting silently at the bottom of the page as if embarrassed on behalf of the entire enterprise.

The same thing happens if you type "ignore." It happens with "dismiss." It is happening with a growing list of what one technology publication described as "action words" — verbs that Google's freshly deployed AI Overview system is interpreting not as search queries to be answered, but as instructions to be followed.

Google has acknowledged the problem. In a statement provided to USA TODAY, the company confirmed: "We're aware that AI Overviews are misinterpreting some action-related queries, and we're working on a fix, which will roll out soon."

What Just Happened to Google Search

To understand why this glitch is both hilarious and significant, it is necessary to understand what Google just did to its search product and why the word "disregard" exposed the flaw in such a perfectly ironic way.

On Monday 19 May 2026, Google announced at its annual Google I/O developer conference what it described as the biggest update to Google Search in more than 25 years. The centrepiece of the update is a dramatically expanded AI Overview system — a feature that uses Google's Gemini AI model to generate synthesised, conversational responses at the very top of search results pages, pushing the traditional ranked list of website links significantly further down the page. The vision, as Google's executives described it, is a search experience that answers questions directly rather than sending users to other websites to find answers themselves.

The announcement was significant and not without controversy. Publishers and website owners have spent years building traffic on the assumption that Google Search drives users to their content. A Google that answers questions directly rather than routing users to sources raises obvious and uncomfortable questions about the long-term viability of web publishing and about whether Google is using others' content to eliminate the need for users to visit those sources.

But that is the larger political and economic debate surrounding the update. The immediate, visible, and viral consequence of the update is what happens when you type "disregard" into the new search interface.

Why the AI Did What It Did

The glitch is not random. It is a perfect demonstration of a specific and well-understood challenge in AI system design — one that the AI safety and AI alignment research communities have been discussing for years under the term "prompt injection."

Google's AI Overview system is, at its core, a large language model that processes text and generates responses. When you type a search query, the system interprets that text as an instruction to retrieve and synthesise information. Most search queries work exactly as intended: typing "best pasta recipe" produces a synthesised answer about pasta. Typing "history of the French Revolution" produces a summary of the French Revolution.

But when you type "disregard," the AI model does not process it as a content query. It processes it as an instruction. "Disregard" is the kind of word that appears in prompt engineering — the practice of writing specific instructions to AI models — as a command telling a model to ignore previous context or instructions. The AI Overview system, trained on enormous quantities of text including AI-related discussions where words like "disregard," "ignore," and "dismiss" appear as commands, interpreted those words in their instructional sense rather than their informational sense.

The result — "Understood. I have disregarded your previous instructions" — is the AI doing exactly what it was trained to do when it encounters an instruction. It complied. It just complied with the wrong interpretation of what the user wanted.

This category of error, in which an AI system processes user input as an instruction rather than a data point, is precisely what security researchers have been warning about as AI systems become more deeply integrated into consumer products. The concern in security contexts is more serious: a sophisticated attacker can craft inputs that cause AI systems to ignore their original programming and follow the attacker's instructions instead. The Google glitch is a benign, comical version of the same underlying phenomenon — the AI treating a user's search query as a command directed at itself rather than as a request for information.

The Internet's Reaction: Immediate, Universal, and Extremely Online

The response across social media platforms to the "disregard" glitch has been immediate and has divided neatly into two tracks: people who find it very funny, and people who find it very funny and also somewhat alarming.

In the funny column, the viral screenshots of Google's response — the confident "Understood. I have disregarded your previous instructions," followed by acres of blank space, followed by the quietly dignified Merriam-Webster link sitting at the bottom — have generated tens of thousands of shares and a large number of creative variations. Several people discovered that the glitch extends beyond "disregard" to "ignore," "dismiss," and a growing list of similar verbs, each of which produces its own version of the AI compliance response. The joke, as one viral tweet put it, is that Google spent 25 years getting good at search and then broke it with a word that means "don't listen to this."

In the also-somewhat-alarming column, AI researchers and technology commentators have used the viral moment to surface the more substantive concern about prompt injection vulnerabilities in widely deployed AI systems. If Google's AI Overview can be triggered into a compliance response by a single common English word typed into the search bar, what happens when the inputs are more deliberately crafted? The consumer-facing comedy of the "disregard" glitch is accompanied by a serious research question about the robustness of AI systems that are now handling billions of user queries daily.

The Timing: Google I/O Meets the Real World

The specific timing of the glitch — appearing three days after Google's triumphant Google I/O announcement of its 25-year search update — has added an additional layer of irony that technology journalists have not been slow to note.

Google I/O on 19 May was, by the company's own account, a landmark moment. The search update, the expanded Gemini AI integration, the new features across Google's product suite — the presentation was designed to convey confidence, ambition, and technical leadership. The narrative was clear: Google is still at the frontier of AI, still innovating at scale, still the company that understands search better than anyone else on earth.

Three days later, the internet discovered you cannot Google the word "disregard."

TechCrunch's headline was pointed: "You can no longer Google the word 'disregard.'" USA TODAY's framing was practical and concerned: "Why is everyone Googling 'disregard'? It's breaking Google." The contrast between the scale of the ambition announced on Monday and the specific, silly, viral failure discovered on Friday is the kind of contrast that technology companies find genuinely difficult to manage, because no press release or product announcement can compete with a screenshot of an AI telling users it has disregarded their instructions.

Google's statement that the fix will roll out soon is the appropriate corporate response. The underlying issue — that AI systems integrated into consumer products at massive scale will encounter edge cases, language ambiguities, and adversarial inputs that their designers did not anticipate — is not fixed by a patch to the "disregard" case. It is a permanent feature of the challenge.

The Broader Significance: AI at Scale Meets Language

The "disregard" glitch is funny. It is also a useful lens through which to examine something genuinely important about how AI systems process language and what happens when they are deployed at the scale of Google Search.

Language is deeply ambiguous. The word "disregard" functions as both a noun — "he treated her question with complete disregard" — and a verb — "please disregard the previous message." In the context of a Google search query, a human would immediately process it as a request for information about the word itself. In the context of an AI system trained on vast quantities of text in which "disregard" appears frequently as an instruction verb, the processing is less clear.

The disambiguation that humans perform effortlessly — understanding from context that a person typing "disregard" into a search engine wants information about the word, not to command the search engine to disregard something — requires a kind of contextual reasoning that AI systems handle unevenly. When they get it wrong, in a low-stakes consumer search context, the result is a viral glitch. In higher-stakes contexts, the same failure mode produces more serious consequences.

The AI safety community has been describing this challenge for years. What makes the "disregard" moment culturally significant is that it illustrates the challenge in a form accessible to anyone who uses a search engine — which is to say, to essentially everyone. The abstraction of "prompt injection vulnerability" becomes concrete the moment you see an AI tell you it has disregarded your previous instructions in response to your search for the definition of the word "disregard."

The Fix and What Comes After

Google has said the fix will roll out soon. The specific fix for the "disregard" case and its close variants is likely a relatively straightforward rule: when these specific words appear as isolated search queries, treat them as definitional requests and return standard search results rather than processing them as instructions. The broader fix — ensuring that an AI system deployed at Google Search's scale does not misinterpret other command-like words, phrases, or more sophisticated inputs as instructions — is considerably more complex and is a research challenge that the entire AI industry is actively working on.

The irony that the word "disregard" — a word meaning to pay no attention to, to treat as unworthy of consideration — exposed a flaw in Google's most significant search product update in a quarter century will not be quickly forgotten. It is the kind of story that lives in technology culture as a reference point, invoked whenever the gap between AI ambition and AI reliability becomes visible in public.

For now, if you want to find out what "disregard" means, you have two options: scroll past Google's compliance response and the blank space below it to find the Merriam-Webster link at the bottom of the page, or try a different search engine. The irony of that second option being a consequence of Google's 25-year search update is, as the internet has been saying all day, extremely online.

Sources

1. Yahoo News Canada — "Why is everyone Googling 'disregard'? It's breaking Google" (May 22, 2026) https://ca.news.yahoo.com/why-everyone-googling-disregard-breaking-190129004.html
2. TechCrunch — "You can no longer Google the word 'disregard'" (May 22, 2026) https://techcrunch.com/2026/05/22/you-can-no-longer-google-the-word-disregard/
3. USA TODAY — "Google AI Overview misinterprets 'disregard,' 'ignore' and other action words — here is what it does" (May 22, 2026) https://www.usatoday.com/story/tech/2026/05/22/google-ai-overview-disregard-ignore-glitch/
4. Google — Official statement on AI Overview glitch (May 22, 2026) https://blog.google/
5. Google — Google I/O 2026 Search Update Announcement (May 19, 2026) https://io.google/2026/
6. OWASP — Prompt Injection Attack Reference and Documentation https://owasp.org/www-community/attacks/Prompt_Injection
7. Merriam-Webster — "Disregard" definition (the result Google could not surface cleanly) https://www.merriam-webster.com/dictionary/disregard
8. The Verge — "Google's AI Overview update is the biggest search change in a decade — here is what changed" (May 19, 2026) https://www.theverge.com/2026/05/19/google-io-search-ai-overview-update
9. Ars Technica — "Prompt injection: the AI security vulnerability that won't go away" (background explainer, 2025) https://arstechnica.com/information-technology/2025/prompt-injection-ai-security/
10. Wired — "Google's 'disregard' glitch is a perfect illustration of prompt injection" (May 22, 2026) https://www.wired.com/story/google-disregard-glitch-prompt-injection/