10. When AI Takes the Content, Where Does the Traffic Go

For more than twenty years the internet ran on a stable supply-and-demand relationship where everyone, or nearly everyone, won.

Sites like blogs gave search engines free content to crawl, and the search engines sent traffic back to those sites. Advertisers could then use that pull to reach potential customers.

In Cloudflare's crawl-to-refer ratio, Google sits at 10:1, OpenAI at 3,700:1, and Anthropic at an absurd 500,000:1. Search engines tend to bring some traffic back. AI agents take the content and send almost none, which amounts to getting it free. That is why many researchers argue the model companies should pay for the human knowledge they use.

For a site that needs visits to keep publishing, a reader who already found the answer elsewhere is a lost visit, a lost subscriber, and a lost chance to be remembered. The content still gets used. The creator may never meet the person using it.

The ratio still needs its measurement caveats. Cloudflare's explanation of the metric compares HTML page crawl requests with refer requests. It does not measure author income directly, and one number cannot tell you what each crawl was for. Cloudflare's 2025 year in review also shows the ratio moving over time.