How the GEO scanner scores

We request /robots.txt at the root of the domain. It counts if it answers 200 with plain text (not with the HTML home page, as many single-page apps do).

We evaluate each user-agent with Google's semantics: the most specific group wins, and within it the longest rule. Bots are weighted by function: search and answer bots (Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot, PerplexityBot, DuckAssistBot) ×3; bots that visit on the user's request (ChatGPT-User, Claude-User, Perplexity-User) ×2; training bots (GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, Meta-ExternalAgent, CCBot) ×1. Blocking a search bot is critical; blocking only training bots is a warning, because it is a legitimate decision.

We request your page three more times, identifying ourselves as GPTBot, ClaudeBot and PerplexityBot (with our signature at the end of the user-agent). If the server answers 4xx, serves a CDN challenge page or returns less than 30% of the text it gave us, we flag it. If the test cannot be completed, we give half the points and say so. If your server rejects even the scanner, we retry once presenting ourselves as a browser so we can analyse the site, and we flag it as a warning.

Half the points for publishing it as plain text; the other half if it follows the proposal: it starts with a Markdown title and has at least 3 links.

The one declared in robots.txt or, if there is none, /sitemap.xml and /sitemap_index.xml. It counts if it contains <urlset> or <sitemapindex>.

At least one application/ld+json block that is valid JSON. A malformed block is critical: engines discard it entirely.

A type that identifies who is behind the site: Organization, LocalBusiness, ProfessionalService, Person and similar.

One point for each piece of entity data: name, url (or @id), logo (or image) and at least one sameAs pointing to external profiles.

A type that describes the page: WebSite, WebPage, Article, BlogPosting, FAQPage, Service, Product, BreadcrumbList, HowTo, among others.

Between 10 and 70 characters: full points. Out of range: 1 point. No title: critical.

Between 50 and 170 characters: full points. Out of range: 1 point.

Declared and pointing to the same site. A canonical to another domain hands over authorship and is flagged.

og:title, og:description and og:image present.

lang attribute on the <html> tag.

Exactly one H1 outside scripts and templates.

No noindex either in the robots/googlebot meta or in the X-Robots-Tag HTTP header (general or aimed at Googlebot or an AI bot).

No nosnippet or max-snippet:0 (critical: you can be found but not quoted). A max-snippet below 50 characters leaves 1 point; data-nosnippet blocks, 2.

Visible words in the initial HTML, outside scripts, styles and <head>. 300 or more: full points; between 150 and 299: half; fewer: critical.

At least two H2/H3.

At least one H2/H3/H4 phrased as a question (“how”, “what”, “why”, “how much”…): the format that best fits how people ask an assistant.

At least one list (2+ items) or table (2+ rows) outside menus, header and footer.

On regular pages: link to “about us” (2 points) and to contact (1). On articles (Article/BlogPosting or og:type article): a declared author, a date within the last 18 months and a link to “about us” or contact, one point each.

The final URL, after redirects, is served over https.

We choose URLs from the sitemap and from the page's links, one per section (/services, /blog, /contact…). On each we check 5 things: that it responds, that it is indexable, that it has a title, JSON-LD and at least 150 words without JavaScript. The score is the share of checks passed. If there is no way to discover other pages, we give half and flag it.