Internal Link Architecture Rebuilt in Your Templates

On a dealer group site with 40,000 stock URLs, roughly 70 percent of internal links are produced by three templates: the global header, the model-family sidebar and the related-vehicles block. Rewriting the rules behind those three moves more link equity in one deployment than a year of editing links inside body copy. That is where this engagement starts and, on most sites, where it ends. How the graph gets built: We crawl the site with Screaming Frog, export the edge list and rebuild the link graph in Python with networkx. We run an internal PageRank over it, then join the result to Search Console impressions by URL and to raw server logs, nginx or Cloudflare, loaded into BigQuery. That gives three columns to read against each other: which URLs the architecture favors, which URLs Googlebot actually requests, and which URLs earn money. The distance between the first and the third is the brief for the rest of the work. What actually changes: Template rules, not links. A related-vehicles block ordered by ORDER BY RAND() is a random walk across your catalogue; ordered by model family, then trim, then dealership proximity, the same block becomes a crawl path with a direction. Each facet parameter gets one written policy, linked and indexable, linked and canonicalized, or not linked at all, and the policy lives in the faceted navigation component rather than being patched in robots.txt afterwards. Pagination, breadcrumbs and the hreflang cluster are treated as part of the same graph, because on multi-market Spanish sites that is usually where the equity leaves. How it is handed over: As tickets your engineers can accept. Each one names the template file, writes out the ordering logic and states an acceptance test. On Next.js, Astro and most headless stacks we will open the pull request ourselves if you want that; on a vendor-locked CMS we write into the vendor's change request format and sit in the call where it is scoped. Nothing here ends at a slide that says improve internal linking. How we prove it worked: We hold about 20 percent of URL clusters out of the change for eight weeks, matched to the treated clusters on template and traffic band, and report the two groups side by side. Internal linking is one of the few disciplines in organic search where a clean control group is available, and almost nobody builds one, which is the honest reason this work stays a matter of taste in the industry. If your release process cannot ship a change to part of a template and not the rest, say so in the first call and we will price the engagement without the holdout. Not included: Content writing. External links. Any change delivered through an edge worker or a JavaScript injection layer. We will not hand you a link architecture that lives inside a tag manager and vanishes the month you stop paying for it. We also do not touch a live faceted navigation without a staging environment we can crawl first. Who this is not for: Sites under about 2,000 URLs. At that size the link graph is legible in a spreadsheet and modelling it is waste. Teams with no route to deploy a template change inside the quarter, because the analysis will be right and will sit unread. And anyone who wants a raised internal link count as the deliverable; the count is not the outcome, and we will argue with you about it. Where we are useful: Dealer groups and marketplaces running live stock feeds, energy retailers with tariff and switching pages, and regulated lenders where legal gates the body copy but not the navigation. Work runs in Spanish and English.