Internal Linking Agent

From a sitemap to a link map

Paste a sitemap URL or a plain list of page URLs. Sitemap index files are followed one level deep, across up to ten child sitemaps. The collected URLs are deduplicated, capped at the first 25, and fetched live in parallel; each page contributes its title, meta description, H1 and the opening stretch of body text. Those signals go to a language model that clusters the set by topic and reasons about which pages are genuinely adjacent, rather than applying generic rules about link counts or exact keyword matches.

What comes back, and what gets checked

Each suggestion names the source page, the destination page, natural anchor text drawn from the destination topic, a one-line rationale and a relevance score from 0 to 100. The server then validates the model rather than trusting it: both URLs must be real pages from the fetched set, a page cannot link to itself, duplicate pairs are dropped, and what survives is sorted by relevance. Orphan pages are derived from the resulting link graph, so any page that received no strong inbound suggestion is listed with a note on how to fix it.

• ›Source page, destination page and suggested anchor text
• ›Relevance score from 0 to 100, strongest first
• ›A one-line rationale for the topical relationship
• ›Orphan pages with no strong inbound suggestion
• ›Counts for pages analysed, pages successfully fetched and links proposed

When to run it

Run it after publishing a cluster, when a new section of the site goes live, or on a site you have just inherited and need to understand quickly. Twenty-five pages is usually enough to tell whether a content set is genuinely interlinked or a pile of isolated posts. Pages that fail to fetch are skipped rather than aborting the run, so a partly broken sitemap still returns usable suggestions, and the response tells you how many URLs were found versus how many were analysed.