Sample: Agent-Readiness Snapshot of blackhull.systems
This is a finished Agent-Readiness Snapshot, made on Blackhull's own website the same way a client's is made.
It covers 14 days of traffic between 12 and 28 October 2026.
01What Blackhull found
- Most "AI agents" were not agents. Of 1,860 requests that named an AI agent, 1,621 were scanners looking for passwords, keys and admin pages.
- Real agents read almost nothing. The 239 genuine agent requests were mostly for the site's rules and map: robots.txt 124 times and the sitemap 90 times.
- AI agents read the pages themselves 21 times in 14 days. None of them read the site's llms.txt, which is written for them.
- Nothing was given away. No scanner request reached anything but a "not found" or a redirect to the front page.
02Who came
- Readers
- ClaudeBot made 152 genuine requests, OAI-SearchBot 46, GPTBot 24 and Amazonbot 10.
- Impostors
- The names Applebot, YouBot, CCBot and DuckAssistBot appeared only on scanner traffic.
- Context
- Across the internet, AI crawlers fetch mostly for training (42.0%) or mixed purposes (43.4%); agents acting for a person right now are 2.3%.
03Five recommendations
- Bring llms.txt up to date. It still describes an offer and terms the site no longer makes.
- Point agents to llms.txt from the places they do read. Agents read robots.txt and the sitemap; neither mentions it.
- Decide the AI-crawler policy on purpose. robots.txt allows every crawler and says nothing about training.
- Answer scanners with "not found", not a redirect. 81 scanner requests got a redirect, which made them look like readers.
- Give agents something to read. Put the services, prices and contact into structured data on each page.
04What a client gets
The same report for their own website, and a 30-minute call.
Contact
AI that runs like an operation, not an experiment. Start with a 30-minute call, no charge.
contact@blackhull.systems