For twenty-five years the web ran on an implicit deal. Publishers let crawlers index their content freely. Search engines sent readers back as traffic. Nobody wrote it down, but both sides needed it.
AI answer engines have ended the deal unilaterally.
What the data shows
Similarweb measured zero-click searches (queries that end without a single click to any website) at 69% by May 2025. By 2026 the figure is above 80%. A Pew Research Center study of 68,000 real queries found that when an AI summary appears in search results, users click an organic result 8% of the time, versus 15% without one. Only 1% of users click the citation links inside the summary itself. Ahrefs measured the click-through rate for content ranked first on informational queries, the kind where knowledge matters most, and found it had fallen from 7.6% to 1.6%. A 79% collapse, for content ranked first.
The other side of the ledger: Cloudflare data puts Anthropic's crawlers at roughly 70,000 page requests for every referral visit sent. The AI system takes, at scale; it returns almost nothing.
A revenue crisis for some
For news sites, travel guides and general media, this is a revenue crisis. Some have closed entirely. "The extinction-level event is already here," said Helen Havlak, publisher of The Verge.
For researchers, the threat is different
For researchers, NGOs and archives, the threat is different, and in some ways sharper.
Their revenue does not primarily depend on web traffic. Their credibility does. Citation and attribution are the currency that academic and mission-driven work runs on — they justify the grant, validate the methodology, establish the institution. The point of publishing a study, a policy paper, or a dataset is not only to inform. It is to be seen as the source that informed.
An AI answer that draws on a body of research without naming it, or listing it in citations that 99% of readers never click, doesn't just lose the visit. It breaks the chain of attribution that makes the work legible as knowledge rather than background noise. A paraphrase without a name attached makes an institution disappear in plain sight, even while its work is actively in use.
What it means to be a canonical source
The question that follows is not only: how do we recover the traffic? It is: in a world where AI systems are the primary interface between knowledge and the reader, how does original work establish that it exists, that it is trustworthy, and that it came from somewhere specific?
Those are not SEO questions. They are questions about what it means to be a canonical source.
Next in this series
Next in this series: what the standards bodies are actually building in response, and how far along they are.