Chapter 2: How AI engines use your blog hub to build answers
Before you build anything, it helps to understand the mechanical process AI engines go through when they turn a user’s question into a recommendation.
Once you see the gears turning, the reason blog hubs work so well becomes obvious.
It also explains why certain types of pages get cited and others get ignored, regardless of how well they rank on Google.
The process has a formal name: Retrieval-Augmented Generation, or RAG. We covered RAG briefly in the first book.
This chapter goes deeper into the specific step that makes blog hubs so powerful, which is query fan-out.
Query fan-out: the mechanism behind everything
When a user types “what’s the best project management tool for a marketing agency?” into ChatGPT, the system doesn’t just search for that exact phrase.
It decomposes the question into multiple sub-queries. Depending on the complexity of the prompt, that decomposition can generate anywhere from 2 to 15 synthetic sub-queries.
For that project management question, the sub-queries might include: “project management tools for agencies 2026,” “marketing agency project management features,” “best PM software collaboration features,” “project management tool pricing comparison,” “Asana vs Monday vs ClickUp for agencies,” and several more variations.
The AI then retrieves web pages for each sub-query separately, scores all the retrieved passages, and synthesizes one unified answer from the best-matching content.
This is the mechanism that makes blog hubs so effective.
If you have a pillar page about project management for agencies, plus cluster pages covering pricing comparisons, feature breakdowns, specific tool reviews, and workflow guides, you’re placing content in front of multiple sub-queries simultaneously.
A competitor with a single blog post on the same topic might catch one of those sub-queries. You’re catching five or six.
Query fan-out: how one question becomes many retrievals
↓
↓
↓ Each sub-query retrieves separate web pages. Your cluster pages can appear for multiple sub-queries. ↓
|
Mike King of iPullRank, who reverse-engineered this process using his Qforia tool, offers the clearest analogy: “The more relevant passages a brand owns across formats, the more chances it has to be cited.
Think of it as accumulating raffle tickets.” A blog hub is a machine for accumulating raffle tickets across the fan-out space.
The raffle ticket model
The raffle ticket metaphor deserves more attention because it changes how you think about content strategy.
In traditional SEO, you pick a keyword, write one page targeting it, and try to rank that single page as high as possible. In AEO, you’re not trying to win one lottery.
You’re trying to have tickets in as many drawings as possible.
Ahrefs’ analysis of over 4 million AI Overview citations confirmed that 31% of cited pages don’t even rank in Google’s top 100 for the original query.
They rank for fan-out sub-queries instead. AirOps found a similar number: 32.9% of cited pages appeared only in results for a sub-query, not the original prompt.
Those pages wouldn’t have been found if the site had only published one article on the topic.
The practical implication: you don’t need to rank #1 for anything to get cited by AI engines.
You need to rank somewhere, for enough related queries, that your content shows up when the AI goes looking.
A blog hub with twelve cluster pages covering different angles of a topic gives you twelve raffle tickets instead of one.
There’s a ceiling, though. Indig’s data shows that only 5% of cited pages get pulled for 10 or more unique prompts.
That top 5% consists almost entirely of broad category-level overview pages (pillar pages) backed by deep cluster support.
The pillar page is your most valuable raffle ticket. The cluster pages are what make it eligible for the drawing in the first place.
The ski-ramp effect and what it means for your content
Kevin Indig’s analysis of ChatGPT citation patterns uncovered something he calls the “ski-ramp effect.” When ChatGPT retrieves a page, it doesn’t weigh all parts equally.
44.2% of citations come from the first 30% of a page. The further down the page you go, the less likely that content is to be cited.
This has direct implications for how you write every page in your blog hub.
The most important facts, the most citable statements, the clearest answers to the reader’s question, all of that needs to go at the top.
If you bury your best material after three paragraphs of introduction and context-setting, the AI may never reach it.
The ski-ramp effect also reinforces why short, focused pages outperform long guides.
On a 5,000-word page, the first 30% is 1,500 words, and the AI may cite something from that section. But the remaining 3,500 words are dead weight from a citation perspective.
On a 1,500-word focused cluster page, the first 30% is 450 words, and that’s where you’ve placed your sharpest, most extractable content. Every word is working harder.
| Citation pattern | Data point | Source |
| Citations from first 30% of page | 44.2% | Kevin Indig / Growth Memo |
| Citations from pages under 1,000 words | 53% | Ahrefs AI Overview study |
| Content length correlation with citation | 0.04 (near zero) | Ahrefs |
| Citation-bearing answers with question headings | 78.4% | AirOps / Indig |
| Proper-noun density in cited passages | ~20% (vs. 5-8% baseline) | AirOps / Indig |
Entity density: why specifics win
The last row of that table deserves its own section. Cited passages have roughly 20% proper-noun density, compared to a 5-8% baseline across all web content.
“Proper-noun density” means the percentage of the text that consists of specific names: people, companies, products, places, organizations, and dates.
Think about what that means in practice. A passage that says “many companies have adopted project management tools to improve efficiency” has zero proper nouns.
A passage that says “Basecamp, founded by Jason Fried in 2004, is used by over 75,000 organizations including Shopify and the University of Chicago” has five proper nouns in one sentence.
The second passage is dramatically more likely to be cited.
AI engines prefer specific, verifiable information because it’s easier to extract and attribute.
When you write for your blog hub, you should be naming products, companies, people, dates, prices, statistics, locations, and standards.
Every fact you add is another anchor the AI can grab onto.
This is one area where blog hubs have a natural advantage over standalone pages. A pillar page can mention many entities at the overview level.
Each cluster page can go deeper on specific entities, adding context, comparisons, and detail that wouldn’t fit on the pillar.
The hub collectively builds a dense entity network that AI engines can map, reference, and cite.
Prompt: Map the fan-out queries for your topicAsk ChatGPT: “If someone asked you ‘[your main topic question]’, what sub-questions would you need to research before answering? List every sub-question you’d look into, including comparisons, pricing, features, use cases, and alternatives.” The sub-questions it lists are your fan-out queries. Each one is a potential cluster page for your hub. |
Prompt: Check your entity densityPaste a section of your existing website content into ChatGPT and ask: “Count the proper nouns in this text (company names, product names, people, dates, locations, specific numbers). What percentage of the text is proper nouns? How could I increase the entity density without making it sound forced?” Aim for 15-20% proper-noun density in your most important pages. |
