Topic clusters were the dominant content architecture from roughly 2019 to 2024. One pillar page covers the broad topic, 5 to 20 supporting posts cover sub-topics, internal links connect them in a hub-and-spoke pattern. The pillar ranks for the head term. The supporting posts rank for long-tail variations. The cluster as a whole signals topical authority to Google.
AI search has not killed topic clusters. It has changed which parts still produce value and which parts are now dead weight. Here is the honest accounting.
The pillar-page concept itself. AI engines preferentially cite pages that read as comprehensive coverage of a topic. The pillar architecture, where one page is the canonical answer to “what is X,” holds up well. Wikipedia’s continued dominance of definitional queries is the strongest evidence: the long-form, structured pillar page is still the dominant citation winner.
Internal linking from supporting posts to the pillar. Internal links continue to matter for SEO and provide weak but positive signal for AI citation. Pages that are well-connected internally are crawled more thoroughly and ranked more reliably.
Topical authority as a domain-level signal. Google still rewards sites that cover a topic deeply versus sites that cover everything shallowly. The cluster as a whole still contributes to this. AI engines partially mirror this through entity binding: a site repeatedly identified as the source on a topic gets cited more often on that topic.
Long-tail keyword targeting per supporting post. The 2019-2024 playbook was to identify 50 to 200 long-tail variations of a topic and write one supporting post per variation. AI Overviews and chat-style answers have collapsed many of these long-tail queries into zero-click answers. The supporting post that targeted “what is the difference between X and Y for use case Z” now mostly serves AI engines rather than human clickers.
The post still has value, but the ROI calculation is different. The supporting post is now more of an entity-and-citation signal than a direct traffic driver. Content shops still pricing supporting posts at the 2022 rate (assuming direct traffic ROI) are pricing them too high.
Cluster size as a goal. The 2022 norm was 50 to 200 supporting posts per pillar. The 2026 reality is that beyond roughly 20 to 30 supporting posts per cluster, the marginal post adds less topical authority than it adds maintenance cost and cannibalization risk. The cluster size sweet spot has shrunk.
Manual internal link maintenance at scale. Sites that built clusters with hand-curated internal linking found that maintaining the links across 100 posts per cluster became a part-time job. Plugins help but rarely substitute. The labor cost of maintaining linking on a large cluster now exceeds the benefit beyond a certain size.
Cluster siblings now compete with each other for AI citation slots.
This is the under-discussed shift. When a cluster has 50 posts on closely-related sub-topics, AI engines have to choose which post to cite for a given query. The engines often choose poorly (citing a tangentially-relevant post instead of the perfect one). The brand’s own posts compete with each other inside the cluster.
The implication: fewer, sharper posts beat more posts when each post is calibrated to a distinct query. The cluster should be designed so each post owns a specific question, not so each post adds keyword density.
Cluster pillars are now also entity signals.
The pillar page is not just the answer to “what is X.” It is the brand’s canonical entity for the topic. Pillar pages should ship with rich Article schema, DefinedTerm schema for the core terms, and explicit links from Organization.knowsAbout to the pillar URL. The pillar becomes part of the brand’s entity binding work.
Supporting posts should be Field Notes or tactical, not just topic coverage.
The supporting posts that still produce value in 2026 are the ones that have a specific point of view, a real example, or a tactical move that requires expertise. The “comprehensive guide to sub-topic” supporting posts have been replaced by AI engines summarizing the topic directly. The supporting posts that survive are the ones with first-person observation or implementation specificity.
Three structural changes from the 2022 default.
One: smaller clusters. 10 to 20 supporting posts per pillar, not 50 to 200.
Each post calibrated to a distinct query. Each post has either tactical specificity (real implementation), original observation (Field Note voice), or comparison value (vs/, alternatives/, three-way comparison). Posts that are just topic coverage do not make the cut.
Two: pillar pages designed as entity hubs.
Full Article schema, DefinedTerm for the core terms, FAQPage for the questions, HowTo if procedural. The pillar should appear in Organization.knowsAbout. The brand becomes identifiable as the source of the topic when the pillar is set up this way.
Three: cluster mapping for AI citation, not just for SERP rank.
The cluster’s purpose has expanded beyond ranking for head terms. The cluster is now the brand’s evidence to AI engines that the brand has depth on a topic. This means the cluster’s quality matters more than the cluster’s size. Three clusters with 15 sharp posts each beat one cluster with 50 mediocre posts.
The audit we run on prospect sites:
Step one: count the clusters. Most sites have implicit clusters they have not formally defined. Identify which posts naturally group together. Anything fewer than 3 clusters is too narrow. Anything more than 8 clusters is probably topically diffuse for a focused brand.
Step two: check the pillars. Does each cluster have a clearly identifiable pillar page? If yes, does the pillar have rich schema, answer-first opening, and recent dateModified? If no, that is the priority gap.
Step three: audit supporting posts for cannibalization. Within each cluster, sample pairs of posts. Do they overlap by more than 30 percent of the outline? If yes, merge or differentiate.
Step four: audit internal linking. Each supporting post should link up to the pillar and across to 2-3 cluster siblings. If the linking is sparse or absent, that is fixable in a single editorial pass.
If your content architecture has not been audited in 12 months, the work is overdue. The 2022 playbook of “more posts, more long-tail” is no longer the right shape. The 2026 playbook is fewer posts, sharper angles, stronger pillars, tighter clusters.
If you want us to map your current clusters and identify the gaps, book the fit call. The audit ships in week one.
Related reading:
– Answer-first writing: before-and-after on real B2B pages
– FAQPage schema is the single biggest AI citation move
– GEO 101: Generative Engine Optimization Explained
– NetPageTwo vs a traditional SEO agency
