A crawler that reaches
an empty page has
still learned nothing.

By September 2026 “AI readiness” had already become a category with a standard shape: a 0–100 score built from three signals — whether robots.txt names and welcomes the AI crawlers, whether an llms.txt file exists, whether the page carries schema.org markup. It is commoditised. Several free, no-signup checkers run exactly that grid, and Apify actors will run it across a site list for $1–5 per thousand pages.

All three of those are permission checks. They tell you whether a crawler is allowed to fetch the page. Not one of them opens the page and looks at what came back.

seo-ai-crawler-access can pass, seo-ai-structured-data can pass, robots.txt can roll out the red carpet for GPTBot, ClaudeBot and PerplexityBot by name — and an AI crawler can still be handed an empty <div id="root"> and nothing else, because the page’s actual content is built by JavaScript the crawler never runs. Permission granted, nothing to read.

rendering-content-requires-js is the check that catches this. It compares the HTML your server sends against the DOM Google’s Lighthouse run produced from the same page — a real render, on Google’s hardware, the only rendered-DOM evidence a scan that launches no browser of its own has access to. When the title, the meta description or the H1 exist only in the rendered column and not the served one, that is exactly what an AI crawler, a search crawler and a link-preview fetcher all see: the served column, and nothing more. This is the same comparison the report shows you as “served HTML vs rendered page,” scored here as part of AI readiness because it is the single most consequential AI-readability signal there is, and the one every permission-only checker misses entirely.

llms.txt (llmstxt.org) proposes a root-level Markdown file summarising a site for a language model, the way robots.txt summarises crawl rules for a crawler. It is a reasonable idea and it is not established: a 2026 measurement across roughly 137,000 domains that had published one found 97% received zero requests from any AI crawler. The one documented real consumer is developer tooling — agents like Cursor or Claude Code reading it when pointed at a codebase or a domain — not a search engine or a chat product answering a question at the moment someone asks it.

A checker that scores this heavily is selling you a checklist item, not a result. Ours grades a missing llms.txt low, on purpose, and says why right there in the finding. Publishing one costs almost nothing and it is not a bad bet on developer tooling — it is just not the thing standing between your site and being read by ChatGPT, Claude or Perplexity. The rendering check above is.