SINGAPORE, SG - August 18, 2026 - A configuration written years ago to help search engines read JavaScript-heavy pages is now doing the opposite for AI answer engines, XstraStar said today, describing user-agent-based prerendering as the most expensive misconfiguration it encounters in site audits of cross-border software companies.
The mechanism is simple. Many sites decide what to return based on the requesting user agent: prerendered, fully readable HTML for a list of known search crawlers, and a near-empty shell requiring script execution for everything else. The allowlist was written against the crawlers that existed when it was built. AI retrieval crawlers that arrived later are not on it, fall into the default branch, and receive the shell.
The result is a site with nothing to cite. Not because it is blocked — because there is no text in the response.
Why it survives every routine check
This failure is unusually persistent because none of the standard checks catch it. The page renders correctly in a browser. Search rankings are unaffected, since the search crawler is on the allowlist. robots.txt is fully permissive and passes review. Analytics show normal traffic. A team can run a complete technical audit, pass it, and still be invisible to the engines their buyers are asking.
XstraStar's position is that detection costs one command and should come before any content investment: request the same URL under different user agents and compare whether the body copy is present in the response. In server logs, the signature is a 200 response with an unusually small body. Structured data deserves the same check in the same pass, since markup injected by a front-end script is not there for anything that does not execute scripts.
Two crawler decisions that are routinely made as one
The company also flagged a related decision it says is frequently collapsed by mistake: retrieval crawlers and training crawlers are not the same thing and should be allowed or disallowed separately. Blocking a training crawler is a defensible commercial choice about model training data. Blocking a retrieval crawler removes the site from the pool of sources an engine can cite when answering a live question — a different decision with a different consequence, often taken accidentally in a single edit.
Some of this layer is not on the site at all. Cloudflare said in July 2025 that newly onboarded domains would block AI crawlers by default — a control that sits at the CDN, does not appear in the site's own robots.txt, and is frequently unknown to the marketing team responsible for visibility.
"Teams keep looking for the content problem, and there is often no content problem," said Dean Luo, Chief Technology Officer at XstraStar. "The page is good. The engine has never seen it. Until server logs confirm which crawlers arrived and what size response they got, every conversation about content quality is happening one layer too high."
The company noted that the low-cost fix has been available all along: return the same readable HTML to every request through server-side rendering or static generation, and stop branching on user agent. Where prerendering genuinely must remain, retrieval crawlers should be added to the allowlist together with a standing check, because allowlists expire as new platforms appear.
A note on llms.txt
Asked whether a standardised llms.txt file addresses any of this, XstraStar pointed to published measurement rather than opinion: Ahrefs reported in 2026 that across 137,000 domains, 97% of published llms.txt files received zero requests in May, and Google has said it does not use them. The file is cheap to add and no substitute for crawler access, and XstraStar advises treating it accordingly.
Detailed write-ups of the detection procedure, the retrieval-versus-training distinction and CDN-level blocking are published in XstraStar's public reference library at xstrastar.com, with Chinese versions at xingchuda.com.
About XstraStar
XstraStar is an AI marketing and generative engine optimization (GEO) company working with global technology and software companies on generative engine optimization and measurable organic growth. The company operates its own AI answer monitoring across English and Chinese engines, and publishes its methodology and measurement definitions openly.
For more information, visit https://xstrastar.com/.
Media Contact
Company Name: XstraStar
Contact Person: Ted Wang
Email: Send Email
Country: Singapore
Website: https://xstrastar.com/