The advice used to be: build silos so Google’s crawler can navigate your site efficiently. That’s still true, but it’s only half the job now. In 2026, I’m not just building for a crawler — I’m building for a synthesizer. ChatGPT, Perplexity, and Google AI Overviews don’t rank pages. They extract from them. If your content isn’t structured to be cited, it doesn’t show up.
I’ve spent the past year deploying content clusters across MyCozyTrove, SafeHarborPrep, HomesAndGardenDecor, OpsForge Labs, and TheCoffeeCan Camp. I’ve watched articles sit dark for months and then surface in an AI summary because I fixed the structural relationship between three related pages. The structure is what changed — not the content.
Here’s how I actually build these clusters.
Why Hub Pages Usually Fail
Most hub pages try to answer everything in one place. That’s the wrong job. An AI system asked “what’s the best solar generator for a CPAP machine?” isn’t going to cite a 5,000-word General Preparedness overview. It’s going to find the specific article that answers that question directly.
The hub page’s only job is to be the map — to tell the search engine that comprehensive coverage exists on this topic and to point toward the specific pages where the details live. It shouldn’t try to be those pages.
The Four-Layer Model
Across all five sites, I use the same cluster structure. It’s not theoretical — it’s literally how the folders are organized in my Git repos.
L1 — The Hub. Broad overview. Links out to the L2 pages. Example from SafeHarborPrep: a guide to off-grid power options that routes readers toward more specific decisions.
L2 — Decision Pages. Category-level pages that help a reader figure out which direction to go. Example: solar versus gas generators for a homestead, or how much backup capacity a household actually needs.
L3 — Trigger Pages. Narrow, problem-specific. These address a specific situation or comparison. Example: how to size a solar array for a well pump. These are where most of the conversion happens — readers arrive already knowing they have a problem.
L4 — Product Reviews. Bottom of the funnel. Specific products, specific use cases.
An L4 review without an L3 to feed it and an L2 to categorize it is an orphan. It exists but it doesn’t belong anywhere, and search engines treat it accordingly. I had several gear reviews on TheCoffeeCan Camp sitting in a loose general category for months with no traction. They weren’t bad articles. They just had no structural context.
What AI Crawlers Are Actually Looking For
AI systems want paragraphs they can lift and use directly. If the first 40 to 60 words of a section don’t contain a direct answer, they move on to the next source.
I tested this on OpsForge Labs with hosting reviews. I stopped using section intros that built toward the point and started putting the answer first. Instead of “Uptime is one of the most critical factors when evaluating a VPS provider,” the section opened with: “This provider averaged 99.98% uptime over a six-month period, with most downtime attributable to scheduled maintenance windows.” That’s citable. The first version isn’t.
Tables help for the same reason. A three-column comparison table is structured data that an AI can parse without having to interpret prose. I use them on every L3 and L4 page now — not because they look good, but because they work.
The Backpack Cluster: What Actually Happened
The clearest test case I have is the backpack cluster on MyCozyTrove. I built 11 interconnected pieces: one L2 hub covering how to choose a backpacking pack, several L3 guides on sizing, ultralight trade-offs, and materials, and six L4 reviews of specific packs.
The linking was bidirectional at publish time. When the L4 reviews went live, they already contained links back to the L3 sizing guide. When the L3 went live, it linked down to the reviews. The search engine didn’t discover a single new article — it found a complete cluster all at once.
Within a few weeks, MyCozyTrove started getting cited in AI results for queries like “how do I fit a 65L backpack.” The citation wasn’t pulling from the review — it was pulling from the sizing guide and then pointing to the review as the next step. That’s the pattern working as intended: the cluster provides the logic, the L4 provides the recommendation, and the AI surfaces both.
Bidirectional Linking Is a Workflow, Not an Afterthought
Most people add internal links after the fact. They publish an article and maybe come back later to link something else to it. That approach means every new article spends weeks as an orphan before it gets connected to anything.
Because I use a Git-based workflow, I treat a cluster as a single push. All the articles are written with their links already in place before any of them go live. When the commit goes up, the crawler finds a fully formed set of relationships immediately — not a single post with promises of future context.
The same discipline applies when adding articles to an existing cluster. When a new spoke goes in, the hub gets updated in the same commit. Not the next day. Same push.
What Changed Once Clusters Were Complete
Before I built out proper clusters, the sites felt like magazines — new content on top, older content gradually buried. Once the clusters were complete, three things shifted.
Time on page went up. Readers landing on an L4 review and realizing they didn’t understand how to size the product would click through to the L3 guide. The cluster was doing what a table of contents does, but contextually.
Indexing got faster. A crawler finding one page in a well-linked cluster finds all of them. I stopped having to manually request indexing in Search Console for new articles on established clusters.
AI citations started appearing. The direct-answer structure within each section gave AI systems something to pull. SafeHarborPrep started showing up in AI overviews for preparedness queries. OpsForge Labs for hosting comparisons. The citations weren’t from the hub pages — they were from the L3 and L4 pages where the specific answers lived.
The Bottom Line
If your articles are sitting in a loose category with no defined relationship to each other, they’re competing against the rest of the internet without any backup. The cluster is what gives them context, and context is what gets you cited.
Answer the question in the first 60 words of every section. Use tables. Link bidirectionally at publish time, not as a cleanup task later.
For how this fits into the overall build sequence, see What I Actually Did to Build Five Affiliate Sites. For how to pick the categories that support these clusters before you build them, see How I Chose Topics and Categories. And for where I got this wrong in practice, Why MyCozyTrove Took Longer Than Expected covers that in detail.