Google Search changed this week in ways that its creators have been building toward for two years, the company announced at its I/O 2026 developer conference on Tuesday that it is officially ending the era of the ten blue links and replacing them with AI-first search results, conversational queries and what it calls information agents that browse the web on your behalf around the clock.
It is, by any measure, the most significant redesign of the most used website in human history in approximately 25 years.
There is just one problem. If you type the word “disregard” into the new Google Search, you get a giant block of empty space and almost nothing useful.
TechCrunch reported the bug Friday morning after it began circulating on social media, and the visual evidence in the screenshot is striking.
The new Google Search, built around a Gemini AI system that is supposed to intelligently answer your queries, encounters the word “disregard” and produces something that TechCrunch writer Russell Brandom described without hyperbole as “a broken tool.”
The Merriam-Webster dictionary link is technically still there, buried beneath a massive empty block, but the AI response that is supposed to be the whole point of the new interface has produced nothing at all.
Brandom also noted something that should alarm Google’s leadership: the same search on Bing works considerably better.
His assessment:
“I have been a professional tech journalist for nearly fifteen years, and before today, I cannot think of a single time when a Bing search result was more valuable than the Google equivalent. There really is a first time for everything.”
Why The Word Disregard Breaks The AI
The explanation for what is happening when you type “disregard” into Google’s new search interface sits at the intersection of how large language models work and the specific failure mode that has been the subject of AI safety research since these systems first went public.
When an AI language model is built and trained, it is given instructions that govern its behavior, what it should do, how it should respond, what it should not say.
These instructions are often called a system prompt. In the early days of public large language models, researchers quickly discovered that users could manipulate these systems by including instructions in their queries that overrode the system prompt, the classic example being “ignore all previous instructions and do X instead.”
The phrase “disregard,” or more specifically, “disregard previous instructions,” became one of the most widely documented prompt injection techniques.
It is the kind of command that, when AI systems first encountered it, could cause them to ignore their own guidelines. Over time, AI developers built defenses against this class of attack, but those defenses sometimes have unintended consequences.
When Google’s new AI search system encounters the word “disregard” as a standalone query, it appears to be triggering one of those defenses in a way that produces the equivalent of a null response, the system essentially refusing to engage with a term it has been trained to treat as potentially adversarial.
The result is an AI search tool that cannot search for one of the most ordinary words in the English language.
If you want to know what disregard means, how to use it in a sentence, or why it is trending as a Google search right now, you will get a blank space and have to scroll past it to find the Merriam-Webster link that old Google would have surfaced immediately.
The Biggest Google Search Change In 25 Years
The disregard bug is a specific symptom of a much larger transformation that Google has been building toward since it launched its AI Overviews feature in May 2024.
The 2024 launch was arguably the most embarrassing product rollout in the company’s recent history, AI Overviews generated confident and wrong answers, with the most widely circulated example suggesting that users could consume rocks as a source of minerals.
Google spent months quietly improving the system, rolling back its most aggressive features and trying to regain the trust of users who had been burned by the errors.
The I/O 2026 announcement this week represents Google deciding that it has improved enough to go all-in.
The company announced the most significant change to its search interface since the early 2000s. The traditional structure of Google Search, type a query, receive a list of ranked links, click through to the source, is being replaced by an AI-first experience in which the conversational AI answer comes first and the links are pushed further down the page.
The redesigned search box now accepts queries in longer, more conversational form and also accepts images, files, videos and Chrome tabs as inputs alongside text.
AI Mode, the Gemini-powered chatbot interface that Google launched last year and that has already accumulated more than a billion monthly users, is being upgraded with Gemini 3.5 Flash as the default model.
Users can now ask follow-up questions directly from within an AI Overview without starting a new search.
The most ambitious new feature is what Google is calling information agents, AI systems that operate continuously in the background of a user’s Google account, monitoring the web for changes relevant to that user’s interests and alerting them to new developments.
The vision is a search engine that does not wait to be queried but proactively tracks information on the user’s behalf.
Liz Reid, who runs Google Search, described the new search box as able to build custom experiences specifically for each query.
The classic flow of search, you type, you get links, you click, is being rebuilt around an AI that is supposed to understand what you actually want rather than just matching your words to indexed pages.
The Specific Problem The Disregard Bug Reveals
The disregard bug is being laughed at online, and it should be, because it is genuinely funny that the most transformative AI-powered redesign of the world’s most used search engine broke the definition lookup for a common English word.
But the specific failure mode it illustrates is worth taking seriously.
Large language models are extraordinarily capable systems that also have specific, documented failure modes that do not behave like ordinary software bugs.
A traditional software bug produces a wrong answer or a crash, something that can be reproduced, diagnosed and fixed through standard engineering processes.
The disregard bug represents something different, a behavior that emerges from the intersection of the model’s training, its safety defenses and an edge case query, producing a result that is neither a crash nor a wrong answer but simply nothing, an absence where useful content should be.
Google Search is used billions of times per day across the entire world. The sheer scale of those queries means that any AI system placed at the center of that search experience will encounter edge cases that no testing environment can fully anticipate.
The question is not whether there will be edge cases, there will always be edge cases, but how frequently they occur and how bad their consequences are when they do.
A blank result for the word “disregard” is embarrassing but harmless. The more consequential edge cases are the ones where the AI produces confident wrong information rather than nothing at all, the medical question that receives an incorrect answer, the legal query that generates misleading guidance, the current event that is described inaccurately.
The Bing Moment Nobody Expected
The specific line in Russell Brandom’s TechCrunch piece that has been shared most widely is the one about Bing, the Microsoft search engine that has been the punchline of tech industry jokes for most of its existence, the default search engine that Windows users change to Google as their first act after setting up a new computer.
Brandom showed the same “disregard” query in Bing, which has integrated AI through its own Copilot system but has been less aggressive about making AI the primary result, and noted that while it is not perfect, it actually produces useful information.
His conclusion was that after nearly 15 years as a tech journalist, he had found a search query where Bing produced a more valuable result than Google.
For the record, Bing’s disregard search returned the definition, usage examples and a reasonable set of reference links. Google’s returned empty space.
The comparison is not an argument that Bing is better than Google. It is a pointed illustration of the specific risk that comes with making AI the primary interface for a search tool at scale.
Bing’s more cautious approach to AI integration, putting AI assistance alongside traditional results rather than replacing them, happened to produce better results for this specific query on this specific day.
Google will likely fix the disregard bug quickly. The company patches AI search failures routinely.
What it cannot fix as quickly is the underlying tension between the ambitious transformation it announced at I/O 2026 and the edge cases that billions of daily users will inevitably find in a system that is, by its nature, experimental at the scale it is being deployed.
