There's a search scenario you've probably encountered.
You open Google, type in a clear query. Then AI Overview gives you an answer—but that answer is either completely blank or entirely unrelated to what you asked.
You search for "Who is the CEO of Company X," and it gives you a primer on "What is a CEO."
You search for a specific number, and it gives you a vague overview.
You search for a person, and it gives you information about someone else.
This isn't an "occasional glitch." This is a systemic fracture.
The Two Faces of AI Search
At I/O 2026, Google unveiled an ambitious blueprint for AI Search: search will no longer just give you links; it will get things done for you. Booking restaurants, comparing products, planning trips—shifting from information retrieval to task execution.
It's compelling.
But there's a massive chasm between flashy features and foundational functionality.
When users search for a specific question and get an irrelevant AI response, they don't think, "This AI will be able to do so much in the future." They just think, "This thing can't even answer basic questions properly."
The Verge's headline used the phrase "so broken." Not "has some issues"—but "broken."
Why This Is Worse Than Traditional Search Errors
Traditional search makes mistakes too. But the nature of "errors" in traditional search versus AI search is fundamentally different.
Traditional search returns a list of links. Users judge for themselves which ones are reliable. If the first link is irrelevant, they click the second. This process is transparent—users can clearly see which results match and which don't.
AI search returns a single answer. A singular, definitive, seemingly authoritative response. Users have no choice. They either accept it or leave.
When AI gives a wrong answer, users are no longer facing "which link is wrong"—but "the answer Google gave me is wrong."
This psychological shift is crucial.
In traditional search, users bear the responsibility of verifying information. In AI search, Google assumes that responsibility—at least, that's what users believe.
So when AI Overview delivers an irrelevant answer, the user's disappointment isn't "the search results are bad"—it's "Google has broken my trust."
The Concept of "Disregarding"
The report uses a heavy word: AI Overviews will "disregard" user queries for certain search terms.
Not "misunderstanding." Not "partially wrong." It's "disregarding."
This means that in certain cases, the AI system isn't even attempting to answer the user's question—it's just generating text that looks relevant but actually isn't.
Technically, this is explainable: large models can trigger flawed reasoning paths for certain queries. But in terms of user experience, it's catastrophic.
A user searches for "side effects of a certain drug" and gets a completely irrelevant response. This isn't a "fun bug"—it could be a safety issue.
Google's Dilemma
Google faces a nearly unsolvable dilemma.
On one hand, they must invest in AI search because competitors (Perplexity, Microsoft Copilot) are already doing it. Not investing means handing over the future of the search market.
On the other hand, the foundational quality issues of AI search remain unresolved. Rolling out AI Overviews widely before large models can reliably answer all types of questions will only create more user disappointment.
But Google can't turn back now. The entire narrative of I/O 2026 is "agentifying search." Stepping back to admit "AI search isn't good enough yet" is commercially and PR-wise impossible.
So Google can only fix problems as it moves forward. Rolling out while patching.
My Take
The problem with Google AI Search isn't a technical issue; it's an expectation management issue.
Google markets AI Search as a smart assistant that "gets things done for you." But what users actually experience is a search tool that might not even answer basic questions correctly.
The gap between these two experiences is exactly how fast trust is eroding.
What's more dangerous is that when AI search fails, its failure mode is far more deceptive than traditional search. A wrong link, users can spot immediately. A wrong AI answer, users might not even realize it's wrong—because its tone is definitive and its format is authoritative.
What Google needs to do isn't "make AI search smarter"—it's "make AI search more honest."
Honesty means: when unsure of an answer, directly say "I'm not sure," instead of fabricating a response that looks relevant but actually isn't.
Honesty means: for certain queries, falling back to traditional search mode and giving users a list of links to judge for themselves.
Honesty means: admitting that AI search is still in a testing phase, rather than packaging it as a fully mature product.
The future of AI search doesn't depend on how many flashy things it can do right now. It depends on whether it can be reliable at "answering basic questions."
If it can't even do that, no amount of agent features will be anything more than castles built on sand.
Primary Sources: