Nobody is clicking AI chatbot source links, Cloudflare CEO warns

AI labs love to boast that their chatbots cite sources. The implicit deal has always been simple: we scrape your words, but in return, we send curious readers straight to your doorstep.

According to Cloudflare CEO Matthew Prince, that deal is officially broken. Speaking to Axios, Prince dropped some brutal numbers showing that referral traffic from AI bots has completely fallen off a cliff.

The brutal referral math

A decade ago, Google sent publishers roughly one visitor for every two web pages it crawled. Six months ago, that slipped to one visitor per six pages. Today, Google crawls 18 pages to return a single visitor.

That looks generous compared to AI chatbots.

Six months ago, OpenAI delivered one visitor for every 250 pages it ingested. Today, it takes 1,500 pages crawled to yield just one click.

Anthropic is in a different league of stingy. Six months ago, its ratio sat at one visitor per 6,000 pages. Today, Anthropic crawls an astonishing 60,000 pages to generate a single referral click.

Users have learned to trust AI summaries. They get the quick answer and move on, completely skipping the footnote links. For publishers who survive on digital ad impressions, that is an existential emergency.

Trapping scrapers in an AI maze

Making matters worse, AI scrapers have routinely ignored the web's honor system. Multiple outlets documented last year that AI companies blithely bypassed standard robots.txt no-crawl instructions to vacuum up training data anyway.

Prince said Cloudflare is now building fresh tools to slam the door on these unauthorized scrapers. And they are not just blocking access; they are actively trolling them.

Back in March, Cloudflare rolled out AI Labyrinth. When an uninvited bot ignores no-crawl directives, the tool serves up an endless web of AI-generated junk pages that look convincing enough to crawl but contain zero real content from the host site.

The scrapers end up burning valuable bandwidth and compute processing hallucinated garbage.

Prince didn't mince words about the showdown: he noted that Cloudflare routinely fends off attacks from state-sponsored hackers across the globe. "And you're telling me, I can't stop some nerd with a C-corporation in Palo Alto?"

Why it matters

This completely obliterates the tech industry's favorite defense: that chatbot footnotes act as a fair trade for scraping the open web. At a ratio of 60,000 to one, the idea of link citation as compensation is dead in the water.

For digital publishers, creators, and website owners, the takeaway is clear: the open web contract is renegotiating itself by force. Relying on search and citation traffic to monetize through banner ads is no longer viable. Publishers must either wall off their content completely or demand direct licensing checks from AI vendors.

We think Cloudflare's technical counterattack is both brilliant and inevitable. When voluntary standards like robots.txt fail, defensive infrastructure takes over. By poisoning scrapers with synthetic decoys and choking off free data, infrastructure providers are shifting leverage back to content owners, forcing frontier labs to finally pay up.