Cloudflare CEO Confirms the Obvious About What A.I. Answer Engines Do to Publisher Traffic ⇥ axios.com

Christine Wang, Axios:

Ten years ago, Google crawled two pages for every visitor it sent a publisher, per [Cloudflare CEO Matthew] Prince.

[…]

Now:

• For Google, it’s 18:1

• For OpenAI, it’s 1,500:1

• For Anthropic, it’s 60,000:1

It is a curious side effect of Cloudflare’s size and position that it is among a true handful of companies with this kind of visibility into a meaningful slice of global web traffic.

In an alternate world, these artificial intelligence businesses may have tried to work with publishers. Perhaps they would have given greater prominence to references, self-policed the amount of summarization they would offer, and provide some kind of financial kickback. Instead, they have trained their systems on publishers’ vast libraries without telling them until it is far too late for it to matter. They take so much while providing so little in return. This will surely accelerate the walling-off of the necessarily paid web, further affirming what I have taken to calling “Robinson’s Law”. This helps explain the increasingly unethical means of acquiring this training data.