106 million ClaudeBot requests: what AI crawlers really cost a specialized media outlet
Anthropic
OpenAI
ActuIA, which allows AI crawlers, reports that ClaudeBot sent over 106 million requests to its servers between February 2025 and May 2026, with a sharp surge from December 2025. A comparative analysis over two weeks showed that Anthropic's crawler generated far fewer referral visits per request than OpenAI's, raising questions about the value returned to content producers.
ActuIA, a specialized AI news media, has published an analysis of its server logs showing that between February 2025 and May 2026, ClaudeBot, Anthropic's AI crawler, sent 106,154,885 requests to its servers. Request volumes were modest until autumn 2025 but exploded starting in December 2025, reaching peaks of over 25 million in April 2026, with ClaudeBot exploring systematically through tags, language versions, and pagination pages, especially after a site redesign. ActuIA never blocked or limited these accesses, with over 99% of requests returning HTTP 200. A two-week observation from May 29 to June 12, 2026, compared crawler requests with referral visitors: Anthropic needed 3,268 requests per visit (57 visits), while OpenAI needed 226 requests per visit (7,018 visits). The heavy crawling consumes bandwidth, computing, disk, and monitoring resources, and complicates audience measurement. ActuIA remains open to AI crawlers, has implemented llms.txt, but argues that openness without compensation is not a viable economic model. It notes that Cloudflare has introduced 'Pay per crawl' and that EU text and data mining rules allow rightsholders to opt out, but a wider industry shift is needed to ensure content producers receive fair value.
- Сокращения
- HTTP = HyperText Transfer Protocol — протокол передачи гипертекста
Source: ActuIA —
original
