The Three Ways a Page Can Be Unreachable

One question, three answers

Ask whether a page on your website is “reachable” and most tools give you a yes. It is in the sitemap. It returns a 200. It is indexed. All true, and none of it answers the question a link-following reader is actually asking: can I walk to this page from somewhere else on the site?

That question turns out to have three different answers, depending on what kind of walking the reader does. Sorting every page into one of those three tiers is one of the core measurements in our industry reports, and this note explains how it works and why the distinction matters more every month.

Two graphs from one crawl

The method starts by building not one link graph of the site but two, from the same crawl.

The full graph keeps every internal link on every page: navigation menus, headers, footers, sidebars, body text, all of it. This is the site as raw HTML presents it.

The content graph keeps only editorial links: the ones an author placed in the body of a page, pointing at another page on purpose. Building it means removing the navigation, and the trick is that you do not need to understand the page design to do that: any link target that appears on most of a site’s pages (we use a threshold of about 80%) is, by definition, site-wide chrome rather than an editorial decision. Drop those, and what remains is the graph of deliberate connections.

The difference between the two graphs is the difference between what a site displays and what it says. A menu that repeats on a thousand pages is real, but it expresses one decision made once, in a template. A link in the third paragraph of an article expresses a judgment about that specific page: this relates to that. The content graph is where a site’s actual knowledge structure lives.

The missing denominator

Both graphs share a blind spot: a crawler that discovers pages by following links can, by construction, only find the pages that links lead to. Whatever nothing points at is not just unranked; it is undiscovered, and its absence is invisible.

That is why the third ingredient is the site’s own inventory: the XML sitemap, the machine-readable list of pages the site claims to publish. The sitemap provides the denominator. Compare the list of pages that exist against the pages the graphs can reach, and the unreachable remainder falls out: countable, listable, exact. Without a usable sitemap the comparison is impossible, which is why our reports mark such sites as not-measurable rather than pretending the answer is zero.

The three tiers

With two graphs and an inventory, every published page lands in exactly one tier.

Tier 1: editorially reachable. The page has at least one inbound link in the content graph. Some author, somewhere on the site, deliberately pointed at it. A reader or agent following in-body links can arrive here.

Tier 2: navigation-only. The page is reachable in the full graph but not the content graph. The only paths to it run through the repeating menu. A human clicking through the navigation finds it easily. A reader that strips boilerplate, which is how most AI agents and content extractors work, precisely because menus repeat on every page and carry no per-page meaning, loses it entirely.

Tier 3: link-invisible. The page is in the sitemap, live, and often perfectly well-built, with zero inbound links in either graph. Nothing on the site acknowledges it. The only ways to arrive are external: a search result, an old bookmark, a direct URL, the sitemap itself.

Why the middle tier is the sneaky one

The invisible tier is the loud failure; once counted, nobody argues with it. The navigation-only tier is quieter and easier to defend: “but it’s in the menu!”, and that defense is exactly the trap. It assumes the reader keeps the menu.

Different readers keep different things. A browser keeps everything. A search crawler keeps most things. An AI agent assembling an answer keeps the content; it discards repeated chrome as noise, because for its purposes it is noise. A site that routes a large share of its pages exclusively through navigation has made those pages conditional: visible to readers that preserve boilerplate, gone for readers that do not. We have measured sites where that conditional share was most of the property: a clean 0% invisible headline wrapped around a site an agent still could not walk.

Reading the tiers as a diagnosis

The three shares, read together, say what kind of publishing habit produced them. A large invisible tier with a healthy core usually means an archive problem: content published, linked once from a rolling index, and forgotten as the index moved on. A large navigation-only tier means a template problem: sections wired into menus but never into prose. A thin editorial tier with both other tiers large means the site was never built for internal linking at all; every page an entrance, no page a hallway.

Each has a different fix, which is the practical reason to measure the tiers separately rather than collapsing them into one “orphan rate.” Re-linking an archive is editorial work on old pages; fixing a navigation-only section is template work on layouts; building an editorial layer where none exists is information architecture. Same symptom to a crawler, three different projects for a web team.

The measurement itself stays simple: two graphs, one inventory, three buckets. It is one of the five readings in the Digital MRI, and the single number we quote most from it, the invisible share, is the answer to a question that no page-level audit ever asks: of everything you publish, how much does your own structure admit exists?