Tested the top 1,000 sites. 40.9% are unreadable to GPTBot. 18.4% are completely dark to every AI crawler tested. The finding that matters most: 17.6% allow GPTBot in robots.txt but return a 403 when GPTBot actually requests the page, because a firewall or CDN is silently blocking traffic the published policy permits. GPTBot is disallowed by 19.1% of sites against 2.8% for Bingbot, a 7x gap.
Caveat worth carrying: this is the top 1,000 sites, which are far more likely to sit behind an aggressive enterprise CDN than a coach on a small host. The 40.9% describes those sites, not a small independent one, and quoting it as if it did would be wrong. What transfers is the failure mode, not the rate.