Chapter 3: Anatomy of a blog hub that actually works
Knowing why blog hubs work is one thing. Building one that actually performs requires getting the structure right: how many pages, how long, how linked, and how they relate to each other.
This chapter lays out the architecture with specific numbers drawn from the 2025-2026 research, so you can plan your hub with confidence before writing a single word.
The three components
Every blog hub has exactly three structural components. Getting any one of them wrong weakens the entire system.
The pillar page is the central node. It covers the broad topic at an overview level, answering the most fundamental questions a reader would have.
It links out to every cluster page, giving the AI engine a map of all the detailed content that supports it.
A pillar page on “email marketing for ecommerce” would cover what email marketing is, why it matters for ecommerce, the major strategies, and brief overviews of each strategy with links to deeper cluster content.
Cluster pages are the depth. Each one takes a single subtopic or question and answers it thoroughly.
Using the email marketing example, cluster pages might cover “abandoned cart email sequences,” “post-purchase email timing,” “email list segmentation for Shopify stores,” “email marketing vs. SMS for ecommerce,” and so on.
Each cluster page should answer one question or cover one narrow topic. If you find yourself covering two distinct questions on one page, split it.
Bidirectional internal links are the connective tissue. Every cluster page links to the pillar. The pillar links to every cluster. And where two cluster pages cover related subtopics, they link to each other.
Yext’s 2025 study found that bidirectional internal linking alone increased citation probability by 2.7 times. Not content quality. Not domain authority. Just the linking pattern.
Linking architecture: pillar, clusters, and cross-links
↓↓↓↓↓
Every link is bidirectional. Clusters cross-link where topics overlap. |
Sizing your hub: how many pages, how many words
The research gives us fairly clear sizing benchmarks. Ahrefs recommends 5-20 subpages per pillar as the practical sweet spot.
HubSpot’s own standard is broader at 20-30 cluster pages per pillar, but that reflects their enormous content operation.
For most businesses, 8-15 cluster pages per hub is a reasonable starting target.
Pillar page length should land between 2,500 and 4,000 words. This is long enough to provide real overview value but short enough that the AI’s ski-ramp effect doesn’t bury your best content.
Remember, 53% of AI Overview citations go to pages under 1,000 words.
Your pillar is the exception because it needs to be comprehensive enough to serve as a genuine overview, but you should still lead with the most citable information.
Cluster pages work best at 1,200-1,800 words of focused, specific content.
That’s enough depth to answer one question thoroughly with supporting evidence, data, and examples, but short enough that virtually all the content falls within the AI’s citation window.
| Component | Recommended Size | Notes |
| Cluster pages per hub | 8-15 (start), up to 20 | Ahrefs recommends 5-20. Start with 8-10 and add more based on performance. |
| Pillar page word count | 2,500-4,000 words | Enough for comprehensive overview. Lead with key facts in the first 400 words. |
| Cluster page word count | 1,200-1,800 words | One question, one thorough answer. All content within the citation window. |
| Internal links per cluster | 3-8 contextual links | 1 to pillar (within first 200-300 words), 2-3 to related clusters, plus any natural product/service links. |
| Maximum click depth | 3 clicks from homepage | Pages within 3 clicks generate roughly 9x more SEO traffic (My Ranking Metrics). |
Internal linking rules that actually matter
Internal linking is where most blog hubs fall apart. People write the content but skip the wiring. Or they add a few links at random and call it done. The research is specific about what works.
Link from every cluster page to the pillar page within the first 200-300 words. Not at the bottom.
Not in a sidebar widget. In the body text, early, with descriptive anchor text that tells both the reader and the AI what the pillar page covers.
“Read our complete guide to email marketing for ecommerce” is better than “click here” or “learn more.”
Link from the pillar page to every cluster page. This is the pillar’s job: to be a directory of all the detailed content in your hub. Each link should use descriptive, keyword-relevant anchor text.
Don’t link everything with the same generic phrase. Cross-link 2-3 related clusters to each other.
If your cluster on “abandoned cart email timing” naturally references a concept covered in your cluster on “email segmentation for Shopify,” link between them.
These lateral links help AI engines understand the relationships between your subtopics and increase the chances that retrieving one cluster page leads to discovering others.
Keep every page in your hub within three clicks of your homepage.
My Ranking Metrics’ research across millions of URLs found that pages within three clicks of the homepage generate roughly nine times more SEO traffic than pages buried deeper.
This matters for AI crawling too. If your hub pages are buried behind four or five navigation layers, AI crawlers may not find them at all.
Why the skyscraper pillar stopped working
The “skyscraper technique,” popularized by Brian Dean at Backlinko around 2015, told content marketers to find the longest article ranking for a keyword, write something even longer and better, and build links to it.
That approach produced pillar pages of 5,000, 8,000, even 15,000 words. For years, it worked on Google.
For AI citation, it doesn’t. Indig and AirOps’ data shows that the skyscraper-style mixed page, the one that adequately answers twenty questions in one long article, is the least reliable format for citation.
The problem is structural: when a page covers twenty subtopics, each one gets a few hundred words.
That’s not enough depth for any one subtopic to score well on cosine similarity when the AI is looking for a specific answer.
Compare that to a focused cluster page that spends 1,500 words answering one question with specific data, examples, and entities.
That page will score much higher on cosine similarity for its specific sub-query, and it will get retrieved where the skyscraper page won’t.
The irony is that the skyscraper technique actually produced proto-hubs. A 10,000-word article with 15 H2 sections is functionally a hub compressed into one page.
The fix is to decompress it: pull each section out into its own cluster page, expand it to full depth, and link everything together.
You end up with more pages, more raffle tickets, more fan-out coverage, and better citation results.
Old model vs. 2026 model: skyscraper vs. focused cluster
|
||
| Exercise: Audit an existing piece of content for hub potential
Pick the longest blog post or guide on your website. Count the H2 sections. For each H2, ask: “Could this section be expanded into a standalone 1,500-word article with its own specific examples, data, and answers?” If the answer is yes for more than four sections, you’ve found a hub hiding inside a single page. In Chapter 4, we’ll map out how to break it apart and build the linking structure. |
||
| Prompt: Evaluate your internal linking
Ask ChatGPT: “I have a blog post about [your topic] that links to [list the pages it links to] and is linked from [list the pages that link to it]. Is this linking structure strong enough for a topic cluster, or are there gaps? What additional internal links would strengthen the hub?” This helps you identify missing connections in an existing content set. |
||
| Prompt: Generate descriptive anchor text
Ask ChatGPT: “I need to create internal links between these blog posts in my content hub: [list your post titles]. For each pair that should be linked, suggest descriptive anchor text that includes the target page’s main keyword. Avoid generic text like ‘click here’ or ‘read more.'” Good anchor text is one of the easiest ways to strengthen your hub’s internal linking. |
