AI Search Is Slowly Strangling the Web's Economic Foundations
The web runs on a simple deal that most people never think about. Publishers create content, search engines send people to it, and those visits generate enough money to justify making more content. It has worked, more or less, for three decades. AI is now quietly dismantling it.
The numbers are starting to come in and they are not pretty. Research from Saharsh Agarwal at the Indian School of Business and Ananya Sen at Carnegie Mellon University found that Google's AI Overviews cut outbound organic clicks by nearly 40 percent while pushing zero-click searches up by 34.5 percent. Sponsored clicks? Completely unaffected. Google gets its ad revenue. Publishers get less traffic.
A zero-click search is exactly what it sounds like: the user asks something, the results page answers it, nobody goes anywhere. The study's conclusion is blunt: AI Overviews redirect traffic away from publishers without any measurable improvement to user experience or engagement quality.
Google's Sundar Pichai has pushed back on this framing, though his arguments have met considerable scepticism, not least because Pew Research data points in the opposite direction.
Alex Chan, an assistant professor at Harvard Business School, has taken a harder look at the structural problem underneath all this. His paper, 'AI and the Collapse of the WWW,' argues that the damage goes well beyond lost page views. Publishers don't just lose the visit and its associated ad revenue. They lose what Chan calls 'durable attention capital': subscribers, repeat readers, backlinks, bookmarks, domain authority, and reputation signals. These are the things that tell future readers and search systems that a source is worth trusting. When AI intercepts the visit, that whole feedback loop breaks.
Chan is careful to avoid catastrophising. He is not predicting every website disappears tomorrow or that all AI homogenises information into grey slop. His actual claim is more precise and arguably more troubling: when an AI platform captures the revenue and the measurement signals without replacing them, the economics of producing quality human content collapse below the point where it makes sense to keep doing it.
Which is, if you think about it, still the web collapse scenario. Just expressed in the language of economics rather than panic.
His proposed fixes are interesting. He is explicitly against a simple 'visitor replacement royalty' where AI companies pay publishers for diverted traffic, and equally against banning AI answers outright. Both approaches, he argues, just prop up a model that was already creaking. Instead, he wants attention to shift toward the new point of engagement, the AI answer itself, while building mechanisms that can actually distinguish expensive, carefully produced human content from cheap AI imitation.
He mentions provenance systems, diversity pricing, exploration credits, and informative audits as tools for restoring the ecosystem. Whether any of that survives contact with reality is another question.
The hard part is that reliably identifying AI-generated content remains genuinely difficult. And there is real money to be made passing off cheap AI output as premium human writing. Anyone hoping for voluntary industry cooperation on labelling should probably reconsider. Where there is profit in slop arbitrage, expect active resistance to anything that threatens it.