No Request With a Known Crawler User-Agent Reached llms.txt: Fifteen Days of Logs at One Origin Server

Many publishers now place a plain-text summary of their site at /llms.txt for language-model crawlers. No standard tells a crawler that such a file exists. This report measures whether crawlers fetched it anyway, using the complete nginx access logs of one origin server for 28 August to 11 September 2026 (15 days, 345,808 parsed requests).

Requests presenting one of 22 published search and AI crawler tokens numbered 46,155 (20 distinct tokens). None of them was for /llms.txt or /llms-full.txt, or for any path ending in llms.txt. From 9 to 11 September, after the files were first served, crawler-token requests numbered 15,906, including 698 for /robots.txt and 282 for /sitemap.xml. All 67 requests for the context files came from clients without a crawler token.

About a third of crawler-token requests bypassed the CDN, and none of those came from the operators' published address ranges, so much of this traffic is probably imitation. The report also corrects the figures printed in Internet-Draft draft-arsentev-llm-context-discovery-00, which were computed on an incomplete log.

Limitations: one origin with three pooled hostnames, about two days of exposure after deployment, origin-only logging behind a CDN, self-asserted identity, and links to the files not recorded. Scripts and aggregates without IP addresses are released with the report.