If your business depends on being the thing people click, Google’s AI Overviews are already changing the math. In a field experiment, Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University found that AI Overviews reduced outbound organic clicks by 39.8 percent and increased zero-click searches by 34.5 percent. A zero-click search is exactly what it sounds like: the user gets an answer right on the search results page, with no need to navigate to a publisher website.

Even more important for the “is this good for users?” debate, the same evidence shows the change did not improve the engagement outcome that publishers can measure. Agarwal and Sen report that the AI Overviews reduced publisher clicks “without affecting sponsored clicks or overall search frequency.” Their conclusion is blunt: the results suggest AI Overviews divert traffic away from publishers “without improving the user experience or quality of engagement for websites.”

This is where the story goes from annoying to existential for the open web. The Register frames the core problem as a new equilibrium that breaks the traditional bargain. As Alex Chan, an assistant professor at Harvard Business School, puts it in his paper “AI and the collapse of the www,” the open web depends on a system: publishers create content, discovery routes users to it, and visits generate revenue. Those visits also create durable information about quality through clicks, subscriptions, and other interactions, which then help future visitors and search systems find better sources.

Generative AI, Chan argues, changes that bargain in a very specific way. An AI answer can draw on publisher content while keeping the user inside the AI interface. “Users may be better served in the short run,” Chan writes, but the source may lose the visit, lose the revenue, and lose the source-level signal that the visit would have produced. That “source-level signal” part matters to boards and strategy teams because it is not just money disappearing; it is measurement disappearing. When measurement and revenue events migrate away from publishers, the feedback loop that rewards human-quality work weakens.

Chan’s scenario is described as a disciplined version of the “AI will make the web collapse” claim. He is not forecasting that every website vanishes, or that AI instantly eliminates information diversity, or that search is doomed forever. Instead, he argues that the risk is structural: when an AI platform diverts revenue and measurement events without replacing them, “costly human information may fall below replacement.” Put plainly, publishers still matter, but the incentives to keep producing costly, high-quality information can fall below what the market is willing to pay.

The Agarwal and Sen field experiment adds empirical weight to the incentive story. Their research focuses on how Google AI Overviews change publisher traffic and user experience, and they tie those effects to the mechanics of zero-click behavior. If AI answers reduce outbound organic clicks by 39.8 percent, publishers lose not only discovery-driven traffic but also the downstream events that often correlate with long-term trust: repeat readership, bookmarks, backlinks, and subscriptions. Chan groups these under “durable attention capital.” His list includes subscribers, repeat readers, backlinks, bookmarks, reputation, and search authority. These are the kinds of signals that, over time, help ecosystems distinguish quality from noise.

The obvious counterargument is that AI answers might benefit users enough to justify the migration. Google CEO Sundar Pichai has argued otherwise, but the Register notes that this view has skepticism around it, “and despite Pew research to the contrary.” That matters because it frames the debate as more than a technical UI change. It becomes a question of whether AI search improves user experience in a way that offsets publisher harm, or whether it simply relocates the interaction where publishers cannot capture it.

So what replaces the missing signals and the missing revenue? Chan is explicit that the solution should not just be penance payments to websites for failing to forward traffic, and he is not interested in a model that props up the old pattern of measurement and discovery. He rejects the idea of a “visitor replacement royalty” and also argues against a ban on AI answers. Instead, his proposal shifts the focus to the new “point of attention,” which for some queries may be the AI answer itself. The goal is to distinguish between costly human information and cheap AI imitation, so the market still knows what to reward.

In Chan’s closing, he calls for “Provenance, diversity prices, exploration credits, and informative audits” to restore the ecosystem: quantity, quality, diversity, source-level signals, and the conventional-search discovery channel that prevents self-reinforcing migration into AI answers. That is easier said than done, especially because distinguishing AI-authored from human-authored content can be challenging. And where there is profit in selling low-cost AI output as premium human-authored content, expect pushback against labeling or other anti-slop mechanisms.

For executives across media, platforms, ad networks, and analytics, the strategic stake is clear. The immediate KPI shift is traffic, but the deeper risk is the erosion of the measurement events that make quality discoverable. The board question is not whether AI answers will exist. It is whether the system will rebuild the economic and informational incentives that keep human expertise funded and verifiable, or whether it slides into a world where attention is captured, signals are lost, and the cost of quality production quietly rises beyond replacement.