OpenAI's Aura safety check can stop ChatGPT and Codex from opening a site

A site that answers 200 to every crawler can still be invisible inside ChatGPT. Search Engine Journal reports that OpenAI runs a second check, called Aura. The check decides whether its browsing tools may open a site at all. OpenAI does not tell the owner when the answer is no.

The article was published on 29 October 2026 by Dan Taylor. It describes what Taylor and other researchers have seen when the OpenAI browsing tools try to open some domains.

What Aura does

Search Engine Journal says Aura is a hostname blocklist that sits in front of OpenAI's agent and browsing tools. It is separate from the GPTBot and OAI-SearchBot crawlers. When an agent tries to open a flagged site, Aura refuses the fetch and returns a short message. The article quotes two of them: "The site-safety policy blocks this action" and "Browser use is not permitted on [domain]."

Taylor writes that the check calls chatgpt.com/backend-api/aura/site_status. A block lasts about 24 hours and overrides the user's own settings and the site's robots.txt. OpenAI does not say in public what gets a site flagged. Taylor writes that security, safety and compliance reasons have been given. He adds that researchers suspect certain prompt histories may trigger a site-wide block.

Why log files do not show it

The block never reaches the website. The crawler sees nothing because the browser never leaves OpenAI's side. The server log stays clean. The owner finds out only when traffic from ChatGPT drops and the site cannot be opened from inside the product.

Search Engine Journal reports there is no self-service page to request a review. There is no formal appeal.

What the article tells owners to check

Taylor lists steps a site owner can run today. Test the domain from ChatGPT's live fetch, from Codex's cloud browser, and from Computer Use. Run the same request against a competitor's domain to tell a site-specific block from a platform-wide outage. Review the server logs for the absence of OAI requests. Confirm that GPTBot and OAI-SearchBot are getting 200 responses. Save the exact error message with timestamps.

If a block is confirmed, Taylor says to open a support case with OpenAI and ask three questions in writing: what the domain's classification is and why, which surfaces are affected, and what the review process is. He advises against workarounds, which violate OpenAI's usage policies.

What this means for a team

AI visibility reports built from crawler logs alone miss this entire class of failure. The crawler was never asked. A real check uses the agent tools themselves. Run it from inside ChatGPT and Codex, on your own domain and on a known-good control. Until OpenAI documents Aura, Search Engine Journal's recipe is the practical test.

Source

OpenAI's Aura Safety Check Can Block Your Site From ChatGPT Browsing, Dan Taylor, Search Engine Journal.