Agent Ready Crawlers

The field guide to every AI crawler and agent at the door, and the vendors who decide who gets in.

How to test what a 403 does to an AI answer about you

In short

Request your own pages with each crawler token and with a plain fetch, compare status codes, then ask the assistant about your site and see whether it quotes you. OpenAI says sites opted out of OAI-SearchBot do not appear in ChatGPT search answers, though they can appear as navigational links.

A 403 to an agent is invisible in your analytics and decisive for the answer. The test takes an afternoon and needs no tool.

In practice

  1. Request your own key pages once with a plain fetch and once with each crawler's user-agent string from the Index, and compare the status codes. Ora counts a 401, 402, 403, 451 or bot challenge to an agent, while a normal browser gets in, as blocked.
  2. Ask the assistant about your site and see whether it quotes you. Ora's lab found (2026-10-19) that serving 403 to agent user agents left both a plain-fetch and a JavaScript client at 0 of 5 retrieved answers on one docs site; that is a vendor finding.
  3. Read the result against what operators say. Sites opted out of OAI-SearchBot do not appear in ChatGPT search answers, though they can appear as navigational links. Disabling Claude-User prevents retrieval in response to user queries.
  4. Remember Perplexity-User generally ignores robots.txt, so a 403 at the WAF is the only refusal it sees.
  5. Compare with the field. Cloudflare reports that less than 1 percent of its sites block search bots while 17 percent block training. The crawl behind this site found 12.9 percent of home pages answering 401 or 403 to its labelled user agent.

What the sources say

Sources

Last reviewed 2026-11-04