There's a certain kind of genius in building a product that destroys its own supply chain while calling it an upgrade. Google has been quietly perfecting this trick for years, but with the rollout of AI Overviews — those cheerful little boxes that now greet you at the top of search results with a confident paragraph summarizing the answer to your question — the company has finally achieved something almost impressive: it has automated the process of making itself useless.
Let's walk through how this works, because the loop is genuinely beautiful in a collapsing-bridge kind of way.
The Answer Machine That Ate Its Own Sources
Google built its dominance on a simple premise: the web has good stuff in it, and Google will help you find it. Publishers created content. Google indexed it, ranked it, and sent users to it. Users clicked. Publishers got traffic. Traffic paid for more content. The flywheel spun.
AI Overviews snap the axle off that flywheel and hand it back to you as a feature.
Now, when you search for something — say, "how long to cook a brisket" or "symptoms of strep throat" — Google's AI reads the top results, produces a tidy little synthesis, and presents it to you before you've had a chance to click anything. You get your answer. You leave. The recipe blog that spent three years building a trusted audience for slow-cooker content gets a visit from absolutely nobody.
This isn't a side effect. This is the product working as designed.
The Irony Is Load-Bearing
Here's where it gets philosophically uncomfortable: AI Overviews are trained, in part, on the very web content they are now systematically defunding.
Think about what that means for a second. Google's model learns from publishers. The model then replaces the need to visit publishers. Publishers lose revenue. Publishers cut staff, reduce output, or shut down entirely. The pool of high-quality content that the model can learn from gets smaller and worse. The model's future outputs get blurrier and less reliable. Google adds a small "Sources" footnote and calls it transparency.
This is not a hypothetical. Traffic data from publishers across categories — health, recipes, finance, local news — has already shown measurable declines since AI Overviews expanded. Some SEO analysts have reported click-through rate drops of 20 to 60 percent on queries where an AI Overview appears. The websites that used to constitute "the web" are learning that Google no longer considers them a destination worth sending anyone to. They are now merely ingredients.
Confident, Wrong, and Uncorrectable
There's another problem that tends to get buried under the traffic discussion, and it's arguably more alarming: AI Overviews are sometimes just wrong. Not in a subtle, nuanced, needs-more-context way. Wrong in a "please do not put glue on your pizza" way. Wrong in a "this medication interacts badly with that one, but sure, go ahead" way.
The model doesn't know what it doesn't know. It produces answers in a register of calm authority that is completely detached from its actual confidence level. A human writer hedges. A human writer links to sources because they know you might want to verify something. The AI Overview simply tells you, in the same tone it would use to describe the boiling point of water, that a certain obscure historical figure did a thing they definitely did not do.
And here's the kicker: because fewer people are clicking through to primary sources, fewer people are in a position to notice the errors. The feedback loop that used to self-correct bad information — somebody reads the original article, spots the mistake, the comments section erupts, the author updates the piece — has been severed. The AI speaks. The web goes unread. The error persists.
Search Solved Itself Into a Dead End
Google's official framing is that AI Overviews make search more helpful. Users get answers faster. Friction is reduced. And on a certain narrow level, this is true. If you want to know the capital of Nebraska, you don't need to click anywhere. Great. Efficient. Fine.
But the web was never just a repository of facts that could be strip-mined and summarized. It was a living ecosystem of opinion, expertise, argument, and discovery. You went looking for one thing and found something better. You followed a link and ended up somewhere you didn't expect. The experience of browsing — actually browsing — rewarded curiosity in ways that a single AI-generated paragraph structurally cannot.
AI Overviews optimize for the transaction. They are extraordinarily bad at everything else.
Who Benefits, Exactly?
The uncomfortable question underneath all of this is: who actually wins here?
Not publishers, clearly. Not users who need reliable information. Not the open web, which loses a little more oxygen every time someone gets their answer from a box and bounces.
Google wins, in the short term, because users spend more time on Google's surface and less time leaving it. Advertisers still pay for the real estate around those AI boxes. The company gets to claim it's innovating. The quarterly numbers look fine.
Until, of course, the content well runs dry. Until the independent blogs and niche publications and local news sites that trained the model in the first place stop existing because they can't pay their writers. Until the AI is left summarizing other AI summaries of content that was itself generated by AI. Until the whole thing collapses into a confident, smoothly written, deeply hollow nothing.
At that point, presumably, Google will announce a new feature to fix it.
The Web Deserved Better Than This
There's a version of AI-assisted search that could have been genuinely good. A version that surfaced sources more prominently, drove more traffic to original reporting, rewarded quality over quantity, and treated the web as a partner rather than a pantry. That version would have required Google to accept slightly less control over the user's journey.
Google did not choose that version.
Instead, we got a product that answers your question, destroys the site that knew the answer, and then asks if you'd like to try Google One for $9.99 a month. The internet is broken. We took notes. Google took everything else.