How the score works

Whether your robots.txt and page directives let AI crawlers fetch, index, and quote the page.

Block the bots that answer questions, or tag the page noindex, and nothing else you do matters. Most sites clear this, so it works as a gate rather than a place to earn points.

scoring — Full marks: every answer-engine and agent crawler allowed, no noindex or nosnippet in robots.txt, the robots meta tag, or the X-Robots-Tag header, on a clean single-hop HTTPS fetch. Blocking a citation crawler or a noindex caps the whole report at 40.

evidence — Opting out of OAI-SearchBot removes you from ChatGPT search results (OpenAI docs).

How much of the real content sits in the raw HTML, before any JavaScript runs.

AI crawlers read the HTML and stop; they don't run JavaScript. Anything injected after load is invisible to them. It's also the hardest pillar to fake, because the only way to pass is to actually server-render.

scoring — Scores the text-to-markup yield and how early the content appears, paired against script weight so an empty shell cannot fake it. A JS-injected mount point caps this near zero; a heavy bundle is fine as long as the text is really in the HTML.

evidence — ~69% of AI-crawler requests execute no JavaScript (Vercel, 2024–26).

Whether the content is structured so a parser can cleanly segment and quote it.

Answer engines strip nav and footer, chunk on heading boundaries, and lift tables almost verbatim. Div-soup with no heading tree flattens into unusable text even when every word is present.

scoring — Graded on heading hierarchy, semantic landmarks, real tables and lists, descriptive link text, and image alt coverage. Presence earns a floor; a clean, chunkable structure earns the rest.

evidence — 44% of ChatGPT citations come from the first 30% of a page, and ~85% of page content is never used (Zyppy, ~1.2M answers).

Whether structured data and metadata state who you are and agree with the visible content.

Engines read JSON-LD facts before prose and use them for the brand and dates shown with a citation. It only helps when it's specific and matches the page. Contradictions cost points.

scoring — Rewards specific, field-complete JSON-LD, a right-sized title and description, OpenGraph, and a self-referential canonical. Generic or thin schema earns little; schema that contradicts the visible content is docked.

evidence — Valid structured data makes a page 2.3× likelier to appear in AI Overviews (Semrush, 2025).

Whether an agent can move past this page to discover the rest of your site.

Reading one page isn't using a site. Agents follow the sitemap and internal links to finish a task. This pillar carries the least weight because most of it is cheap to add, and the much-hyped llms.txt is still ignored by answer engines.

scoring — Rewards a real sitemap.xml, a dense graph of unique internal links, breadcrumbs, and a genuinely curated llms.txt. Declared files are capped so they cannot buy the score.

evidence — 97% of llms.txt files received zero AI requests in 2026 (Ahrefs).