OpenAI's hidden Aura safety check can block sites from ChatGPT browsing
OpenAI runs a central site-safety check, referred to internally as Aura, that decides whether its browsing agents may connect to a website at all. A flagged site is refused by ChatGPT's live search and Codex's Browser Use before any request reaches the server, and the block overrides both robots.txt permissions and a user's own allow rules. There is no public way to see why a site was blocked or to appeal.
The check is separate from OpenAI's published crawlers. GPTBot, OAI-SearchBot and ChatGPT-User follow standard web rules and publish their IP addresses, but the safety layer sits in front of them and can stop an agent from opening a link regardless. Network traffic shows status checks going to chatgpt.com/backend-api/aura/site_status, and a blocked verdict is cached locally for 24 hours.
One international media publisher with open robots.txt rules and support for OpenAI's tools has been unable to load its pages in ChatGPT since May 2026, with server logs showing no agent traffic arriving at all. OpenAI support escalated the ticket and confirmed in writing that a safety block was active and under review, without naming the Aura system. The publisher's training-crawler traffic was also thin, around 5,000 hits in 30 days against 477,000 from Google and 564,000 from Anthropic.
For brands chasing citations in AI answers, a block at this layer removes the page from the conversation entirely, so no amount of structured content or entity work will surface it. The practical route is diagnostic: test the domain across ChatGPT and Codex, confirm server logs show no incoming agent requests, rule out your own firewalls, then take that evidence to OpenAI support and keep following up.
Reported by Search Engine Journal ↗
Topics: ChatGPT, Citations, Search policy