cg-web-seo

Keyword density, word-count targets, “the meta description is a ranking factor”, “a sitemap entry means indexing” — the skill lists these beliefs as forbidden. Every recommendation is labeled as a documented guideline, a technical best practice, an empirical heuristic, a tool convention or a hypothesis.

Claude builds a topic map first and a keyword map second. Every primary search intent gets one clear target page, and a new URL is created only when it serves a distinct user need. Near-identical landing pages and city-swapped variants are ruled out.

Crawling, indexing and canonicalization are kept apart, because they are different mechanisms. robots.txt controls crawling, noindex controls indexing, a canonical is a signal rather than a command, and a soft 404 is caught by comparing what the user sees with the status code the server sends.

JSON-LD as a connected graph with stable entity IDs, limited to what is visible on the page and supported by the search engine. Schema is treated as help with understanding and rich result eligibility — never as a ranking booster.

The skill lists sentences Claude must never say, and replaces them with probabilistic, testable language:

From being found to being measured. A page that cannot be crawled never reaches relevance, and a page that ranks but does not convert has not finished the chain.