Most AI-visibility advice starts with structured content, entity clarity, and robots.txt. But a new layer sits in front of all of it: OpenAI’s central site-safety check, internally called Aura. If your domain lands on that blocklist, ChatGPT’s live fetch and Codex’s Browser Use will refuse to open your pages before a single request reaches your server.
Why robots.txt doesn’t save you
The crawlers OpenAI publishes—GPTBot, OAI-SearchBot, and ChatGPT-User—follow standard web rules, but they are separate from the policy layer that governs browsing agents. When Aura has a site classified as blocked, an explicit allow rule on the user side or in your own setup won’t override it. Network traffic shows status checks hitting chatgpt.com/backend-api/aura/site_status, and a blocked decision is cached for 24 hours.
OpenAI publicly says it blocks sites for safety, security, and compliance reasons. That makes sense. The problem is there is no public status page, appeal form, or explanation for site owners when a legitimate business is caught by mistake.
Evidence from a real publisher
Search Engine Journal’s investigation followed an international media publisher that ran a paid campaign with OpenAI and kept robots.txt open. Since May 2026, ChatGPT repeatedly failed to load its pages, while server logs showed zero inbound requests. In controlled tests, competitor news sites loaded in the same session, a third-party page running the same technology on a different domain loaded, and Chrome permissions set to “Always allow” made no difference.
The error was blunt: “The site-safety policy blocks this action.” Weeks later, all five tested URLs were still blocked. The publisher logged roughly 5,000 successful OpenAI crawler connections in 30 days, compared with 477,000 from Google and 564,000 from Anthropic.
How to check whether it’s you or OpenAI
- Ask ChatGPT to open a specific URL, then try Codex Browser Use; a block usually appears across surfaces.
- In the same chat, open two or three competitor sites. If they load and you don’t, it’s not an outage.
- Check server logs: an OpenAI-side block leaves no request, while your own firewall leaves a 403 or challenge.
- Test subdomains and a page on another domain running your tech; hostname-level classification blocks subdomains but not the third-party domain.
- Record the exact error and timestamp before opening a ticket.
What to do while there is no appeal button
Build an evidence pack: screenshots, log extracts, crawler analytics, and robots.txt. Open a support case, keep the case number, and ask explicitly whether your domain has a site-safety classification, which surfaces it affects, and what the review process is. Use account contacts if you have them. Don’t try to bypass the block; OpenAI’s own agents are instructed not to circumvent the policy, and attempting workarounds can turn a false positive into a justified restriction.
Why this matters commercially
If ChatGPT’s browsing agents can’t open your pages, your site stops existing in AI search regardless of your GEO or AEO investment. Agents simply choose an accessible competitor. ChatGPT then answers from secondary sources, forums, or stale content, which is a reputation risk for brands that depend on accuracy. The worst part: because the block happens before a request leaves OpenAI, your analytics may never show the loss.
Source: Search Engine Journal
