A senior content architect's playbook for designing content hubs that earn topical authority.

Most content programs accumulate accidentally. Pieces ship one at a time, internal linking happens (or does not) in editing, similar topics get covered by multiple pieces with no coordination, and six months later the team has a content set that looks like a sitemap but feels like a yard sale. Search engines see disconnected pages; AI engines see scattered citations; readers cannot tell which piece is the entry point and which is the reference.

This skill is the discipline of intentional hub architecture. It assumes you have decided which topic areas to invest in (see ) and have keyword data (see ). It does not write any individual piece (see for per-piece work and for execution). What it does is design the structure: which topic gets a pillar, which subtopics get cluster pieces, how the pieces link to each other, where they live in the URL hierarchy, and what topical authority signals the architecture is engineered to produce.

When to use this skill: launching a new content hub, restructuring an existing content set into a hub, deciding whether a piece should be a pillar or a cluster, or auditing why a content area is not ranking despite individual piece quality.

This skill spans hub-level content architecture. It composes with five sister and adjacent skills, and the distinction between them is what keeps each one sharp.

The clean composition: decides which topics, surfaces opportunities, this skill designs the hub, briefs each piece, writes each piece, later catches what slipped. Six skills, sequential.

The audience: SEO content strategists, content architects, in-house teams designing content hubs, agencies running topical authority programs. The voice is senior content architect to junior content marketer. Concrete, opinionated, honest about what makes a hub earn authority versus decorate the sitemap.

The keystone distinction. Three categories of content in any program.

Pillar. Comprehensive piece covering a major topic area. 2,500 to 5,000 words plus. Anchors the topical authority claim for the topic. Receives links from cluster pieces; links out to cluster pieces selectively, not as a footer dump. The pillar is the entry point for the topic and the reference the cluster orbits.

Cluster. Narrower piece covering a specific facet of the pillar topic. 800 to 2,000 words. Links up to pillar; may link sideways to other clusters in the same hub when a connection is natural. Each cluster answers one specific question within the pillar's scope.

Orphan. Piece that does not connect to a pillar or cluster. May rank or convert on its own merits, but does not compound into topical authority. Some orphans are fine by design (release-note pages, customer stories, one-off campaigns); most orphans are accidents.

The pathology. Most content programs are 80% orphans, 15% accidental clusters, 5% accidental pillars. The discipline is moving toward 70% intentional clusters, 20% intentional pillars, 10% standalone-by-design. Same total content volume; different architectural outcome.

The reading model. Search engines and AI engines both read internal-link graphs to infer topical authority. A pillar with 12 well-linked cluster pieces signals depth on a topic. 12 orphans on the same topic signal scattered, less-authoritative coverage. The architecture is the signal.

Detail and decision tree in

references/pillar-cluster-decision.md

.

Not every topic deserves a pillar. The five-criterion selection framework:

The "everything is a pillar" anti-pattern. Teams call any 3,000-word piece a pillar. Length does not make a pillar. The architecture does. Pillar is the role a piece plays in the hub structure, not the word count.

Detail in

references/topic-selection-criteria.md

.

Once a pillar is selected, plan its cluster:

Detail in

references/cluster-planning-patterns.md

.

The hub's internal-link graph is what actually produces topical authority signals. The architecture has three directions:

Pillar to cluster (top-down). Pillar links out to each cluster piece via contextual anchors, typically once per cluster from within the pillar's body. Do not bury cluster links in a "related reading" footer; weave them into the relevant section of the pillar where the reader's curiosity peaks.

Cluster to pillar (bottom-up). Every cluster piece links up to the pillar at least once, typically in the first 200 words and again in the closing section. Bottom-up is the discipline that makes the pillar compound; without it the pillar is just a long article.

Cluster to cluster (lateral). Selective lateral linking when one cluster naturally references another. Do not force lateral links; an "everyone links to everyone" cluster is messy and dilutes anchor text relevance.

Anchor text discipline. Vary the anchor text. "Click here" is dead; exact-match keyword stuffing is dead; descriptive natural-language anchors are alive. The brief for each piece (see ) specifies anchor text per outbound link.

The "linking inventory" pattern. Maintain a sheet (or database row, or dbt model) tracking every link in the hub, who links to whom, and the anchor text used. Audit quarterly to catch broken links, anchor-text drift, and missing connections that should exist.

Detail in

references/internal-linking-architecture.md

.

URL patterns matter for both crawl clarity and reader navigation. Decision factors:

Subfolder vs subdomain. Subfolder is the default () for SEO consolidation; both pillar and cluster live on the same domain authority. Subdomain is appropriate only for genuinely distinct properties (, ).

Hub URL pattern. for the pillar, for clusters. Avoid if the goal is hub authority; the URL itself signals topical grouping to crawlers and readers.

Slug conventions. Short, descriptive, keyword-aware but not stuffed. not .

Breadcrumbs. Surface the hub structure in breadcrumb navigation. Helps readers see where they are; helps crawlers see the hierarchy.

The "/blog/ trap." Teams put pillar and clusters under because that is where the CMS folder is. This works structurally but loses the URL signal that says "these pieces belong together as a topical hub."

Detail in

references/url-structure-patterns.md

.

A pillar page does specific work. The standard sections:

Pillar length. 3,000 to 5,000 words is the typical range. Pillar quality is comprehensiveness AND depth, not just word count. A 6,000-word pillar that covers 3 facets is worse than a 4,000-word pillar that covers 12 facets.

Detail in

references/pillar-page-anatomy.md

.

Cluster pieces have a different shape:

Cluster length. 800 to 2,000 words typical. A cluster piece longer than 2,500 words is probably a mini-pillar that should either get promoted to its own pillar or be split into two cluster pieces.

Detail in

references/cluster-piece-anatomy.md

.

Signals that make hub architecture compound into authority. The two-engine view: pillars and clusters serve search engines and answer engines together, not as competing optimizations.

The two-engine optimization is complementary, not competing. Both reward depth, structure, and entity coverage. The hub architecture creates more surface area for both signal types than scattered orphan content can.

Detail in

references/topical-authority-signals.md

and the skill for AEO/GEO strategy at the program level.

Hubs decay if not maintained. The discipline:

The "set and forget" failure. Hub launches, ranks, drives traffic for 2 years, then decays without anyone noticing because nobody owns the hub long-term. Hub ownership is durable: single owner across 3 to 5 year horizons, not "the team that launched it."

Detail in

references/content-refresh-patterns.md

.

Rapid-fire. Diagnoses in

references/common-pillar-failures.md

.

When designing or auditing a content hub, walk these 12 considerations.

The output of the framework is an architecture document the team can reference at every stage: pillar selected, cluster planned, links specified, URLs chosen, page anatomies templated, refresh cadence set, owner named.

Comprehensive content can be replicated. Authoritative sources can be cited. Individual pieces can be matched in quality by competitors with similar resources. What is harder to replicate is the compounding of an intentional hub built and maintained over years: the link graph, the entity coverage, the maintenance discipline, the topical depth signals that build over a 3 to 5 year horizon.

That is the moat. Architecture is the only part of content marketing that compounds reliably; everything else is execution that can be matched.

When in doubt about whether a hub is ready, ask: is the pillar selected against the five criteria, are the 10 to 12 clusters planned with facet heterogeneity, is the link graph specified top-down and bottom-up, is the URL structure clean, is the pillar anatomy templated with TL;DR and FAQ, is the refresh cadence set, and is the durable owner named? If yes to all of those, ship the hub plan and let the per-piece briefs and writers work.