Pillar and cluster: the architecture Google now rewards

For fifteen years, the dominant content strategy in SEO was keyword first, page second. Find a high-volume query. Write a page targeting that query. Repeat for every query you care about. The output was a flat heap of pages, each engineered to rank for one term, loosely organized by topic in the navigation.

That strategy is being quietly retired by Google’s ranking systems — and by the way generative models retrieve content. The architecture that replaces it isn’t new. It’s a refinement of how academic publishing has worked for decades. We call it pillar and cluster.

I want to walk through what the architecture is, why it now outperforms the flat-keyword model, and how to actually build it without ending up with a Wikipedia clone.

What pillar and cluster means

A pillar page is a comprehensive resource on a broad topic. Not a blog post. Not a 1,200-word essay. A real reference — the kind of page someone would bookmark and refer back to. Think “answer engine optimization, complete guide” or “choosing a contractor in [your metro], everything to know”. Three to seven thousand words is normal. Multiple sections. Internal navigation. The pillar is the table of contents for an entire topic.

A cluster page is a focused, narrower answer to a specific question that the pillar’s topic raises. “How long does AEO take to show results” or “licensed vs unlicensed contractors, what’s the actual difference”. Eight hundred to fifteen hundred words. One job: answer one specific question completely.

The clusters link up to the pillar. The pillar links down to each cluster. The internal-link graph for the topic resembles a wheel — the pillar at the hub, the clusters as spokes.

A topic with one pillar and ten to fifteen clusters is a topic cluster. A site with five topic clusters has fifty to eighty pages, all topically coherent, all cross-linked. That’s the architecture.

Why this beats flat-keyword publishing

Two reasons, both grounded in how ranking actually works in 2026.

First, Google’s ranking systems have moved from keyword-document matching to topical authority matching. The question isn’t anymore does this document contain the keyword the user typed. The question is does this document’s domain demonstrate authoritative coverage of the topic the keyword belongs to. A flat heap of pages, each targeting one keyword, doesn’t accumulate topical authority. A topic cluster does. The pillar and its clusters reinforce each other; each cluster is evidence that the domain knows the broader topic deeply.

Second, generative models extract better answers from clustered content than from flat content. When ChatGPT or Perplexity browses a page during retrieval, the model is looking for a clean, structured answer it can lift. A cluster page targeting one specific question, with a clear answer in the first paragraph and structured detail below, is exactly the format the model can extract from. A flat keyword-targeted page, padded out to “rank” for multiple terms, gives the model nothing clean to pull.

The same architecture that wins traditional Google rankings also wins AI citations. That coincidence isn’t coincidence — it’s the convergence of how both surfaces have learned to evaluate content.

What goes in a pillar

A pillar earns its bookmark by being the page a reader could send to a colleague and not be embarrassed about. The structure that does this consistently:

A short, sharp lede that states the topic and why someone should care. One paragraph, no fluff.

A table of contents linking to every section. This is non-negotiable. It tells the reader the page is a reference, not a sales piece. It also lets AI crawlers parse the page’s structure cleanly.

Each major section: a clear heading, a one-paragraph summary of the section’s argument, then the detailed explanation. The summary paragraph is what AI assistants will lift when they cite the section — it should be self-contained and useful even out of context.

Cross-links to clusters wherever a section raises a question the cluster page answers in detail. “For more on how AI Overviews actually rank citations, see our deeper piece on retrieval mechanics.”

A small, restrained call to action at the bottom — typically a single sentence with a single link, not a sidebar.

What goes in a cluster

A cluster page is single-purpose. It answers one specific question. Its structure:

A title that is the exact question, phrased the way a buyer would type it. Not “AEO timeline”. “How long does answer engine optimization take to show results”.

The answer in the first paragraph, complete. The reader who scrolls no further should have what they came for.

Three to five paragraphs of detail, examples, caveats, edge cases. This is where the page earns its rank and its citation — a model needs structured detail to extract a good summary, not just a one-line answer.

A clear back-link to the parent pillar at the bottom. “This question is part of our broader guide to answer engine optimization.”

FAQPage schema tagging where appropriate. If the page has a clear Q&A structure, mark it as such in JSON-LD. AI assistants are dramatically more likely to surface a Q&A pair when the schema announces it.

How to actually build it

The mistake I see most often is teams treating pillars as a one-time project — “we’ll do the pillar this quarter, then the clusters over the next two”. That doesn’t work. The pillar by itself ranks for nothing. The clusters by themselves rank for individual queries but don’t accumulate authority.

The correct build order is parallel: ship the pillar in week one with three clusters already linked. Ship two more clusters every week for the next two months. By the end of month three, the topic cluster has fifteen pages, the pillar is comprehensive, and the internal-link graph is producing compounding gains on every page.

Then start the next topic cluster.

Five topic clusters, each with a pillar and ten to fifteen clusters, is a complete content architecture for a typical service business. That’s seventy-five to eighty pages. At a thirty-article-a-month cadence, a business builds this in roughly three months.

The brands that build the architecture have a structural advantage that flat-keyword publishers cannot match without restructuring their existing pages. Which most flat-keyword publishers will never do, because the rewrite is more work than the original publish was.

That gap — the one between the brands that built the architecture and the brands that didn’t — is going to define which businesses still rank in 2028 and which ones have been quietly demoted out of the SERPs they thought they owned.

---

Luke LaFave is the founder of LaFave Consulting. The studio’s monthly content program is built around the pillar-and-cluster architecture described here.