OpenAI’s Agents Can Silently Block Your Site – & Your Logs Won’t Show It
If ChatGPT's agents quietly fail to reach your pages, you may be missing from AI answers without knowing why. Server logs alone won't show it.
On this page
Allowing OpenAI's crawlers in robots.txt does not mean its browsing agents can open your site
Search Engine Journal reported in late October 2026 that OpenAI runs a central site-safety check, called Aura in its systems and codebase, that decides whether its browsing tools may open a given URL. If a domain is flagged, Codex's Browser Use and ChatGPT's live search refuse to load it, whatever the site's robots.txt says. The author found the check in network traffic, where requests go to a site_status endpoint on chatgpt.com. According to the piece, the blocked verdict is cached locally. There is no public way to learn why a site was flagged or to request a review. The author says established companies, including well-known media brands, have been caught by it.
What is the Aura block, and how is it different from OpenAI's crawlers?
It is a separate safety gate for autonomous browsing tools. GPTBot, OAI-SearchBot and ChatGPT-User are the documented crawlers, and they follow robots.txt. Aura sits outside those rules.
Most publishers' AI-visibility work stops at the crawler layer: allow the right user agents, publish IP ranges, and check the boxes. The SEJ piece argues that layer is not the only gate. OpenAI's help pages do say agents can be kept from some sites for security, safety and compliance reasons using a global blocklist. They do not explain how the list is built. Our earlier note on a shared AI training opt-out covered the consent side of this. Aura is a different mechanism, and no opt-in or opt-out signal from you reaches it.
Why won't my server logs show the block?
The check runs on OpenAI's side before any connection to your server. A refused fetch leaves no request for you to log, so your analytics show nothing.
That is our reading of the mechanism the article describes, not something it states in so many words. The consequence is that a blocked site looks the same as a site nobody asked about. The author says an outage in the status check defaults to allowing access, but an explicit block holds. It holds even against allow rules an enterprise customer sets up internally. So the usual debugging steps of checking logs, loosening robots.txt and whitelisting IPs will not tell you anything about this failure.
How would I know if my domain is flagged?
Only by testing from the agent side. Ask ChatGPT's live search or Browser Use to open your pages and record the refusals. The server side gives you no evidence.
The article's pitch is that you should gather test results and logs to support a request for review, but it also says no public review route exists today. Treat that gap as the main point. The cause of a flag is unknown. The author notes that the status API receives chat details along with links, and some researchers suspect certain prompt histories may trigger site-wide blocks. That is a hypothesis, and the article does not present it as confirmed.
What should I ask OpenAI or my AI-visibility vendor?
Ask whether the vendor's "AI crawlability" check includes agent-side fetch tests, and ask OpenAI directly for a documented appeal path and the criteria behind the blocklist.
Be sceptical of audits that report only robots.txt status and crawler hits. They can pass a site that agents cannot open. Also note what the article does not give: any count of affected domains. "Most websites never run into these restrictions" is the author's impression, not a measured rate, and the evidence here comes from one researcher's reading of traffic and code, not from OpenAI. If you sell through ChatGPT-driven shopping or rely on live-search citations, one repeatable refusal is worth escalating through your account contact.
Frequently asked questions
Does a permissive robots.txt protect me?
No. According to the report, a flagged site is refused regardless of robots.txt or local user settings.
Is Aura the same thing as Atlas?
No. The author notes that Aura here is not the earlier codename for Atlas.
Is there an official appeal process?
Not publicly. The article says there is currently no public way to learn why a site was blocked or to request a proper review.
Source: Search Engine Journal
The original story
Identify OpenAI Aura blocks that stop agents from opening your site, and gather evidence to request a review. The post OpenAI’s Agents Can Silently Block Your Site – & Your Logs Won’t Show It appeared first on Search Engine Journal.