Topic clusters and internal linking: the content architecture that compounds
Most content programmes stall because they publish articles, not architecture. The teams that pull away treat their site as a structured graph of intent, where every page makes its neighbours stronger and authority compounds instead of leaking. That structure is the difference between a blog that ranks for nothing in particular and a domain that owns a subject.
The pillar/cluster model, done as intent mapping rather than keyword stuffing
The pillar-and-cluster model gets butchered more often than any other framework in content marketing. The common version is: pick a fat head keyword, write a 4,000-word "ultimate guide", then bolt on a dozen articles targeting variations of the same phrase and link them all back. That produces keyword cannibalisation, not topical authority. You end up with eight pages all chasing "project management software", competing with each other for the same SERP slot, and a pillar that reads like a glossary nobody finishes.
Done properly, a cluster is a map of intent, not a thesaurus of a single keyword. The pillar owns the broad, mostly informational or commercial-investigation intent for a topic. The cluster pages each own a distinct sub-intent: a specific question, a comparison, a use case, an objection, a how-to. The test for whether two pages belong as separate cluster members is simple: would a reasonable searcher be satisfied by one page answering both queries, or do they expect two different answers? "How much does X cost" and "Is X worth it for a small team" feel adjacent, but they are different jobs and deserve different pages. "X pricing" and "X cost" are the same job and must be one page.
I build clusters by starting from the decision a buyer is making, not from a keyword tool export. I'll sketch the actual sequence of questions someone asks on the way from "I have a problem" to "I'll buy this", then attach search demand to each step. Keyword volume tells you how to phrase a page and whether to build it at all; it does not tell you what the page should exist to do. When you invert that order, you get a cluster that mirrors how people actually research, which is also, not coincidentally, what search engines and language models are trying to reconstruct.
Internal linking is the mechanism, not the decoration
If the cluster is the architecture, internal linking is the load-bearing structure. It does three jobs at once, and most sites only think about one of them.
- It distributes authority (PageRank-style link equity). Links pass ranking signal. Pages that receive more internal links from relevant, authoritative pages tend to rank better. Your pillars should be among the most internally-linked pages on the domain, because they are the ones you want to rank for the highest-value, highest-competition terms.
- It defines topical relationships. The pattern of links between pages tells a crawler which pages are about related things. A tight web of links among your cluster pages, plus links up to the pillar and across to sibling clusters where genuinely relevant, signals "this site has depth on this subject" far more reliably than any single long page.
- It controls crawl and discovery. Orphan pages (pages with no internal links pointing at them) get crawled rarely and rank poorly. Links are how new content gets found and how importance gets communicated. If a page matters, it should be reachable in a few clicks from your most-linked pages.
The diagnostic I run first on any established site is an orphan and depth audit. Crawl the site, then cross-reference the crawl against the URLs in the XML sitemap and in server logs or analytics. Any page that exists, earns traffic or revenue, but is orphaned or sits five-plus clicks from the homepage is a structural failure you can fix in an afternoon. On larger sites I'll quantify it: I want the great majority of indexable, valuable pages within three clicks of the homepage, and I want zero orphaned money pages. That single fix routinely moves rankings before a word of new content is written, because you are redistributing authority that was already trapped.
Information architecture and URL structure: the skeleton has to match the map
Your URL and folder structure should be a legible expression of the cluster map, because both users and crawlers read structure as meaning. If your topical model says a subject has three pillars and a dozen clusters, the site's navigation and URLs should make that hierarchy obvious. I generally favour shallow, descriptive, lowercase paths that group cluster content under a clear parent, for example:
/guides/technical-seo/ <- pillar
/guides/technical-seo/crawl-budget/ <- cluster page
/guides/technical-seo/log-file-analysis/
/guides/technical-seo/canonicalisation/
I have a strong bias toward keeping URLs stable. URL structure carries history: links, rankings, and crawl signals all attach to a specific string. Restructuring for its own sake is one of the most common self-inflicted traffic losses I see. If you must move URLs, do it with 301 redirects mapped one-to-one, never with a blanket redirect to the homepage, and expect a re-crawl and re-evaluation period. The trade-off is real: a cleaner architecture can be worth the temporary turbulence, but only when the existing structure is genuinely confusing crawlers, not merely offending an internal sense of tidiness.
Breadcrumbs deserve a mention here because they are an underused, low-effort lever. A breadcrumb trail with BreadcrumbList structured data reinforces the hierarchy for crawlers, gives every cluster page a contextual link up to its pillar, and improves the way the page is represented in results. It is internal linking and information architecture expressed in one component.
Anchor text: descriptive, varied, and never engineered
Anchor text is a genuine ranking signal for internal links, and it is also the place where well-intentioned teams over-optimise themselves into trouble. The principle is straightforward: internal anchors should describe the destination in natural language. If a cluster page is about reducing crawl budget waste, linking to it with the anchor "reducing crawl budget waste" is good. Linking to it forty times across the site with the exact phrase "best crawl budget optimisation tool" because that is your target keyword is a footprint, and it reads as manipulation to both readers and ranking systems.
My working rules for internal anchors:
- Make the anchor describe what the reader gets, in the words they would use. Vary it naturally across pages, the way a human writer would, rather than repeating one optimised string.
- Avoid generic anchors like "click here" and "read more" for links that matter; they pass weaker topical signal and help nobody.
- Place links in body content where they are contextually earned, not just stuffed into a "related posts" footer module. An in-content link surrounded by relevant text carries more weight and more user value than a templated block.
- Keep exact-match commercial anchors for internal links rare and natural. You control your own internal links completely, so an unnatural exact-match pattern is entirely self-inflicted and entirely avoidable.
One nuance worth holding onto: when multiple links on a page point to the same destination, search engines have historically tended to weight the first link's anchor. So if your navigation already links to a pillar with a short label and your body text links to it again with a richer descriptive anchor, do not assume the descriptive one is the one being counted. It is a small thing, but it argues for getting your most descriptive, intent-rich link to a page as early in the primary content as you reasonably can.
Pruning and consolidating thin content: subtraction is a growth tactic
Almost every content programme older than two years is carrying dead weight: thin posts that never ranked, near-duplicate articles competing for the same query, seasonal one-offs, and "we needed to publish something this week" filler. This matters more than people expect, because site-level quality signals are real. A domain stuffed with low-value, low-engagement pages can suppress the performance of its good pages. Pruning is therefore not housekeeping; it is a way to concentrate authority and quality signals onto the pages that earn.
My consolidation process is deliberately unsentimental. For every URL, I look at organic traffic and rankings over a meaningful window, conversions or assisted conversions, backlinks earned, and crawl frequency, then sort each page into one of four buckets:
- Keep and improve. Earns traffic or links, fits the cluster map. Strengthen it and its internal links.
- Consolidate. Two or more pages competing for the same intent. Merge the best content into one canonical winner,
301the others into it, and reclaim their backlinks and rankings in the process. - Update or re-target. Right topic, stale or thin execution. Rewrite against current intent rather than deleting.
- Remove. No traffic, no links, no strategic value, off-topic for the domain. Either
301to the nearest relevant page if it has any equity, or let it return410 Goneif it has none and you want it out of the index cleanly.
The threshold I apply for "no value" is not a single magic number, but a pattern: negligible organic clicks over twelve months, no ranking keyword in a striking-distance position, no referring domains worth preserving, and no role in the conversion path. When a page fails all of those, deleting or merging it almost always helps the cluster around it. On a publisher site carrying years of archived posts, a disciplined prune that consolidated dozens of overlapping articles into a smaller set of authoritative pages is one of the more reliable ways I know to lift the survivors without writing anything new.
How this builds topical authority and entity strength
Everything above ladders up to two related ideas that increasingly drive how both classic search and AI systems decide what to surface: topical authority and entity strength. Topical authority is the accumulated signal that your domain comprehensively, credibly covers a subject. You earn it by covering a topic with real depth across an interlinked set of pages that map to genuine intent, not by publishing one long article and hoping. The cluster model is, in effect, the production system for topical authority: it forces breadth and depth at once and wires them together.
Entity strength is the adjacent concept. Modern search is built on a knowledge graph of entities, things with defined identities and relationships, rather than just strings of keywords. When your content consistently describes the same entities (your product, the concepts in your field, the people behind the work) and connects them through clear internal links and structured data, you make those entities legible. You are helping the machine build an accurate model of who you are and what you are an authority on. Schema markup, consistent naming, an author and organisation entity that is described the same way everywhere, and internal links that connect related entities all feed this.
This is also where my own work outside SEO reinforces the point. I build local-first AI software and publish technical books, and the throughline is the same: a body of work that is internally consistent, deeply interlinked, and clearly attributed to a single identity is what lets any system, search engine or language model, treat you as a known quantity rather than an anonymous string of pages. Authority is structural before it is editorial.
Why this architecture is now the price of entry for AI search
The shift to AI-driven answers (AI Overviews, assistant-style results, and chat interfaces that synthesise rather than list) does not make this architecture less important. It makes it the price of entry. Retrieval-augmented systems work by pulling relevant passages, attributing them to sources, and synthesising an answer. They reward exactly the properties a well-built cluster has: clearly-scoped pages that answer a specific intent in a self-contained way, unambiguous entities, and a coherent site that demonstrably knows the subject.
There are concrete implications for how you build. Self-contained, well-structured passages, clear headings that pose and answer real questions, and definitions stated plainly, get extracted and cited more readily than meandering prose. The cluster map gives an AI system multiple confirming signals that your domain is a credible source on a topic rather than a single page that happened to match. And entity consistency is what lets a model attribute a claim to you with confidence. Practically, the same things that make a cluster rank well, intent-true pages, strong internal linking, structured data, pruned-away noise, are the things that make your content retrievable and citable.
The mistake would be to treat "optimising for AI search" as a separate discipline requiring a separate playbook. It is the same architecture, held to a slightly higher standard of clarity and self-containment. If anything, AI search punishes the keyword-stuffed pillar and the thin filler post more harshly than classic ranking ever did, because there is no ten-blue-links consolation prize; you are either part of the synthesised answer or you are invisible.
Building the system that keeps compounding
The reason I call this architecture rather than tactics is that, done well, it compounds. Each new cluster page strengthens its pillar; each pillar strengthens the domain's topical authority; that authority makes the next page easier to rank and easier to cite; pruning keeps the quality signal high as you scale. The work is in the system, not the individual article. Map intent honestly, link with purpose and restraint, structure the site so it reads as a coherent body of knowledge, cut what dilutes it, and keep your entities consistent. Do that for a few years and you stop chasing keywords one at a time. You own the subject, and ownership is what keeps paying out long after the publish date.
Where this fits in my work
This is the kind of technical-SEO and growth work I ship end to end, not just advise on. You can see the full portfolio of sites, software and publications I’ve built, browse what I do, request my SEO strategy and content architecture services, or get in touch about applying it to your site. Related reading: Getting cited by AI: AEO and GEO and Technical SEO that actually moves revenue.