Internal link architecture: what links to what, and why

Internal linking earns a budget line on one condition, and we test it in the first week: your important pages are already being crawled and still lose to weaker ones. Concretely, if more than 15 percent of your revenue-carrying URLs sit five or more clicks from a hub, and your access logs show Googlebot reaching them less than once a month, there is money in the structure. If neither is true, we say so, invoice the diagnostic and stop. That has happened three times since 2023. Who decides: Hanna Löfgren leads this work. Six years in-house at a Nordic retail group, where she decided which of roughly 400,000 product and guide URLs kept their links across three replatforms, and where she twice argued successfully for deleting a whole content section rather than linking into it. She signs every recommendation. No junior analyst decides what links to what, and no model does either. The question the engagement answers: Given your crawl, your logs and your revenue by URL, which pages should send links, which should receive them, and which should stop being linked at all. The third part is what gets argued about in the review call, because it usually names pages somebody on the marketing team wrote and defended in a previous job. How the answer is reached: Full crawl in Screaming Frog or Sitebulb, 90 days of raw access logs, Search Console query-by-URL exports, and your own revenue or pipeline figures per URL. Pages are grouped by the queries they already earn impressions for, not by folder or template, because folder structure is a publishing convenience and it lies about intent. We then compare the internal link weight each destination receives against the revenue it carries, and rank the gap. A link is proposed only where the source page already ranks in the top 20 for a query the destination should own. A link from a page nothing reaches passes nothing on. What is delivered: A change set as CSV and JSON: source URL, destination URL, anchor text, the module or paragraph it belongs in, and one sentence of reasoning per row. A separate removal file, sequenced to ship after the additions. A rules document for your template layer, so the pattern holds for the pages we never saw. And a one-page brief for whoever writes the ticket, in language an engineer can estimate from without us in the room. Not included: We do not write body copy, and we do not implement in your CMS. We would rather brief your developers than guess at your components from the outside. No external link acquisition, no redirect mapping, no hreflang. If one of those turns out to be the binding constraint, we will say so and stop billing for link work until it is resolved. When the finding is unwelcome: In 2024 a fashion client with a seven-figure organic line was told their linking was adequate and that they were losing category terms to two marketplaces on price and delivery display, not on structure. That went in the first paragraph of the summary, not a footnote. They did not buy the follow-on work. We would write it the same way again. Who should not buy this: Sites under about 2,000 URLs, where an afternoon with a spreadsheet gets you most of the value. Anyone who wants a tool run and a score, which Ahrefs will sell you for a fraction of this. And anyone whose engineering team has no capacity this quarter, because a change set that waits six months in a backlog is describing a site that no longer exists, and we will not regenerate it for free. What is at stake: Where the diagnostic passes, the pages we redirect weight toward are usually already sitting in positions 6 to 15, the band where a two-position move shows up in the revenue line rather than the impressions chart. That is why the work is priced against your organic revenue, and why we would rather decline four estates than take one where structure is not the problem.