Topical authority is a network property, not a page property

The single-page fallacy

Teams still brief one article at a time. The pattern is depressingly consistent: monthly editorial calendar, twelve titles, twelve briefs, twelve writers, twelve publish dates. The result is a portfolio of unrelated pages that share a domain but nothing else.

Search systems increasingly reward the opposite: dense clusters that cover an entity from every angle, with internal links using descriptive anchors, and a pillar page that serves as the canonical entry point. A cluster with twelve pages will out-perform twelve unrelated pages on the same domain by a wide margin, even when the individual writing quality is identical.

The reason is that ranking systems are increasingly modelling topics as neighbourhoods in an entity graph. A page's authority on a topic is derived not just from its own signals but from the signals of the pages it links to and receives links from within its own domain. A lone page is a data point; a cluster is a neighbourhood.

What a cluster actually is

A cluster is a pillar page plus 8–20 supporting pages, all internally linked with descriptive anchors, all covering a single entity or category. Not tags. Not categories. A designed graph.

The pillar answers the broad question ('what is X, why does it matter, when do you use it') at the depth of a knowledgeable overview — long enough to be authoritative, short enough to still be readable. The supporting pages answer the narrow questions that fall out of the broad one, one question per page, each linked from the pillar and each linking back to it.

A well-built cluster reads like a small textbook on the topic. A poorly-built one reads like a tag page. The difference is intent: the cluster is designed as a knowledge unit; the tag page is a byproduct of taxonomy.

Designing an entity graph

Start with the entity, not the keyword. Map its facets, its adjacent entities, and its user intents. That map becomes the content plan. For an HVAC client, the entity might be 'AC repair'; the facets are the failure modes (compressor, refrigerant, capacitor, thermostat, drainage); the adjacent entities are the systems it touches (ductwork, insulation, thermostats); the intents are diagnostic, transactional, and preventative.

That single exercise generates 20–30 briefs organically. Compare that to keyword-first planning, which generates 20–30 briefs that share a keyword theme but no underlying structure. The cluster-first plan will out-rank the keyword-first plan within two quarters, in every one of our tracked engagements.

This is where the 90-day content roadmap in a typical engagement comes from. We spend the first two weeks on entity mapping and the last ten weeks executing against the map. Very little is invented mid-quarter, which keeps the writing team fast and the SEO consistent.

How we build clusters in engagements

Pillar first, supporting pages in waves of 4–6, internal linking pass after every wave, republishing pass every quarter. The rhythm is deliberate: pillar sets the anchor, wave one covers the highest-intent supporting questions, wave two covers adjacent facets, wave three fills in gaps that emerge from analytics after waves one and two ship.

The internal linking pass is where most agencies stop and most of the compounding happens. Every new supporting page gets links from every other supporting page it references, plus a link to and from the pillar, plus reciprocal links from any related cluster that shares an entity. This is deeply unglamorous work; it's also where the graph structure actually gets built.

We run the linking pass in a spreadsheet with anchor text, source URL, and destination URL logged. When we return to the cluster in six months we know exactly what got linked from where, which makes the republishing pass fast.

Internal linking as compounding infrastructure

The internal link is the most under-priced asset in SEO. A single descriptive internal link from a high-authority page to a new supporting page routinely gets that supporting page indexed and ranking within days, at essentially zero marginal cost. The same lift from an external link would cost hundreds of dollars in outreach time.

The compounding property matters more than the individual lift. As a cluster matures, every new page adds internal-link paths to every existing page. A twenty-page cluster with dense internal linking effectively re-shares authority every time a new page is added, which is why cluster performance tends to accelerate rather than plateau — provided the linking discipline holds.

The counter-example: clusters where new pages are added without updating the internal links from existing pages. The new pages ship, they get indexed slowly, they never accumulate authority, and the cluster stops compounding. The pages are fine. The graph is broken.

The republishing pass nobody runs

Every quarter we walk back through every cluster we've built and republish. Republish means: refresh statistics and screenshots, rewrite paragraphs that no longer reflect current practice, add sections addressing questions that have emerged from user behaviour, update internal links to include any new pages from adjacent clusters, and — this is the underrated part — resurface the page in navigation and internal linking as if it were newly published.

The lift is reliable and boring: 15–40% traffic increase on republished pages within 30 days, holding for two quarters or more. The reason we call it underrated is that most teams treat 'evergreen' as 'set and forget' when the actual play is 'evergreen means republished on a schedule.'

The republishing pass is also where AI-search citation persistence lives. Pages that get refreshed on a visible cadence hold citations across model updates far longer than pages that ship once and go stale.

Common failure modes

Publishing supporting pages before the pillar. The supporting pages have nowhere to link to and no canonical authority anchor. Fix: pillar always ships first, even if it's shorter than final.

Using tag pages as pillars. Tags are auto-generated; pillars are designed. A tag page cannot serve as a pillar because it lacks the editorial argument that makes a pillar authoritative.

Skipping the internal-link pass because it's tedious. This is the mistake that kills more clusters than any other. If you cannot commit to the linking pass, don't commit to the cluster.

Treating the cluster as done at launch. A cluster is a living system. If it isn't in the quarterly republish rotation, it will slowly decay into a tag page's worth of authority.