Semantic Clusters Form From Links, Not From Pillar Pages

Most sites have content on similar topics. Very few have structure.

The difference is not how much you publish. It is whether your pages teach anything to the systems crawling them.

A site can have fifty articles on SEO and still look, to a search engine, like fifty strangers in a room. The question is not "do we have internal links?" It is: do our links explain how our pages relate to each other?

Semantic clusters are the answer to that question. Not because they organize your content calendar, but because they make every linking decision deliberate.

What a semantic cluster actually is

A semantic cluster is a group of pages that collectively cover a topic with enough depth and internal coherence that a search engine can recognize the site as an authority, not just a collection of posts.

The most common way to describe a cluster is "pillar plus spokes." A broad pillar page is planned at the center, and deeper spoke pages are written to orbit it.

We see it the other way around. A cluster isn't drawn in advance. It forms when pages close in meaning link to each other in their text, and the pillar is simply the page that ends up receiving the most of those links.

A cluster on content marketing might include an article about editorial calendars, another about repurposing content, another about measuring performance. Each of those pages is close to the others in meaning, not just in subject.

A page is read with the pages around it, not alone.

A single well-written article on internal linking is a good page. That same article, sitting inside a coherent cluster of pages on link building, SEO architecture, and site structure, is a signal. Search engines see not just the page, but the web of related content around it.

Two ways to get a cluster

A brief history most SEOs don't know

French consultant Laurent Bourrelly formalized the cocon sémantique methodology years before HubSpot popularized "topic clusters" at Inbound 2017. Bourrelly's approach was built on semantic proximity in linking, not just content grouping. The goal was to wire pages together based on how close they were in meaning, so that crawlers traversing the link graph would perceive a coherent thematic structure.

Sylvain and Guillaume Peyronnet later refined this into an explicitly algorithmic version. Their Cocon SEO approach uses semantic vector proximity to calculate which pages should link to which, at scale.

The French SEO community has been doing algorithmically what HubSpot turned into a brand framework. The anglophone world got the name. The French got the methodology.

What we keep from it is the principle of linking pages by how close they are in meaning. The planned tree, with a target page at the top, is the part we leave aside.

How search engines read a cluster

When Googlebot crawls your site, it doesn't evaluate pages in isolation. It follows links, maps relationships, and infers which pages belong together. The cluster structure you build determines what it concludes.

A site that demonstrates comprehensive, interlinked coverage of a subject signals something fundamentally different from a site that has one well-optimized page.

The crawl logic works like this: a crawler lands on a page, reads the content, then follows outbound links to understand what the page considers related. If those links consistently point to pages within the same semantic territory, with anchors that describe the destination accurately, the crawler builds a map of topical coherence.

If those links are scattered, the map stays blurry.

This is where most sites lose the signal. A link in a menu tells a crawler a page exists. A link woven into the body of an argument tells a crawler what that page means.

The anchor text is how a search engine learns what the destination covers. An anchor like "click here" teaches nothing. An anchor like "flat site architecture" tells the crawler exactly what it will find on the other side.

A 2025 analysis of 2.5 million contextual internal links across 1,700 real websites (LinkStorm) found that nearly half of in-body links show minimal semantic overlap between the anchor text and the target page's topic. Most "contextual" links are navigational links placed in body copy. They check a box without building a signal.

Here is what the difference looks like in practice:

❌ Bolted on (the link is detached from the argument): "Internal linking is an important part of any SEO strategy. Read our guide to site architecture to learn more."

✓ Woven in (the anchor is part of the sentence's meaning): "Pages buried more than three clicks from the homepage rarely accumulate enough internal link equity to compete for competitive terms, which is why flat site architecture is the structural prerequisite for any cluster strategy."

Same destination. The second link teaches the crawler what "flat site architecture" means in relation to the surrounding argument. The first just opens a door.

A study of 23 million internal links (Zyppy) found that pages with 40-44 internal links received four times more organic traffic than pages with 0-4. But after roughly 45-50 links, traffic declined. More links is not more signal. More meaningful links is.

Clusters form when every page is linked both ways to its neighbours

You don't need to finish one cluster before starting the next. Link each page, in and out, to the pages closest to it in meaning, and clusters take shape across the whole site, with hubs emerging from the links.

The planned approach picks a topic, builds its pillar, then completes every spoke before moving on. It is tidy on paper, but it treats a site as a series of projects rather than a body of content that keeps growing.

We work page by page instead. Each page receives links from the pages that discuss its subject, and gives links to the pages its argument leads to.

Even one link in each direction shows up in the data. In our study of 1,804 blog articles, an article that receives at least one contextual link from another article and gives at least one ranked for about 56% more keywords than an isolated article, at comparable length, age and backlinks.

Hubs then appear without being designated. A broad article on internal linking is the natural reference for every narrower page on anchors, depth or audits, so it collects links from all of them.

A planned pillar is a bet on which page should lead a topic. The links record which page actually does, page after page, as the site grows.

A cluster is never finished. It is healthy when every page in it receives at least one contextual link and gives at least one, and when its broad articles collect links from the narrower ones.

When clusters connect, and when they should

A cluster is not a prison. When a page genuinely addresses something that overlaps with another topic on your site, linking across clusters is the right call. The only question worth asking: does this link serve the reader's next logical question? If yes, it belongs. If it's a stretch, it doesn't.

Some pages belong to two clusters simultaneously. A page on "measuring content marketing ROI" might sit at the intersection of a content strategy cluster and an analytics cluster. That page is not an anomaly: it is a bridge page. It earns links from both clusters and links back to the broad articles of each. This is a feature of a mature site architecture, not a problem to fix.

The distinction worth making: a bridge page genuinely covers a topic that sits between two clusters. It is not a page that has been artificially stretched to reach a second cluster for the sake of cross-linking.

Your SEO architecture cluster might naturally link to your content strategy cluster. A page on internal linking structures could reasonably link to a page on editorial planning. That link is not a dilution. It is a bridge, and it strengthens both clusters by showing the search engine that these topics are genuinely related.

The test is editorial, not algorithmic. Would a reader who just finished that paragraph naturally want to know more about that other topic?

✓ "A page buried four clicks deep receives a thinner share of the site's value, which is why crawl depth matters as much to a cluster as the links inside it." A natural bridge between two clusters.

❌ "Internal linking matters for SEO. See our pricing page to get started." A navigation link masquerading as a contextual one.

Well-placed inter-cluster links are how a site's authority network expands. They are not a contradiction of cluster thinking. They are its mature expression.

AI search reads the same links, one passage at a time

AI answer engines retrieve passages, not pages. A cluster where each page answers one question and links to its neighbours with descriptive anchors gives them passages that are easy to find and to place in context.

Most AI systems use a retrieval mechanism called RAG (Retrieval-Augmented Generation). The system pulls the most relevant passages from indexed sources, assembles them into context, and generates an answer. It does not read your full site. It reads specific chunks of specific pages.

The unit of visibility is no longer the page. It is the paragraph.

This changes what "well-structured content" means in practice. A page that thoroughly answers one specific question, then links to related questions with descriptive anchors, becomes a self-contained citable unit.

A page that tries to cover an entire topic in one go is harder to chunk, harder to retrieve, and harder to cite.

Our own data ties these links to AI citations. Across 1,804 blog articles, at comparable length, age and backlinks, the share of articles cited in Google's AI Overviews varied with their contextual links.

• 35% of articles that receive and give at least one contextual link were cited at least once.
• 22% of isolated articles were.

That is an association rather than proof of an effect. It still points the same way as the retrieval mechanics, and the work is the same for both kinds of search.

What this looks like in practice

Building a semantic cluster is not a content sprint. It is a linking habit, applied to every page you publish and to the ones already online.

1. Link every new page both ways. The day it goes live, link out from it to the pages its argument leads to. Then add links to it in the older pages that discuss its subject, because the links that help a new post most come from older ones.

2. Reconnect the pages already isolated. List the pages that receive no contextual link from another page, or give none, and add one in each direction. An internal link audit that counts contextual links in both directions finds them.

3. Let the hubs show, then confirm them. Look at which broad pages collect the most contextual links from narrower ones. Those are your hubs, and when you decide what belongs in the menu and what belongs in the text, they are the first candidates for a menu slot.

Two questions per page. Does at least one page close in meaning link to it in its text? Does it link to at least one in return? A page that fails either check is partially isolated. It exists, but contributes little to the cluster.

Signs a cluster is working vs. signs it needs attention