2.3× more AI citations from granular structure

Granular, audience-specific pages get cited by AI 2.3 times more often than generic pages that cover the same topic broadly. That's the finding from an analysis of AI citation patterns by ALM Corp, an internationally active marketing agency — and the number lands on a sore spot I see on almost every site I review technically: too many topics squeezed onto a few broad pages, instead of a few topics spread across many sharply scoped ones. That single decision is what determines whether an AI cites you — or a more tightly focused competitor page.

What the analysis actually shows

ALM Corp compared how often audience-specific pages — built around a particular company size, role, industry or use case — get cited for targeted queries, versus generic product pages trying to cover everything at once. Result: 2.3 times more citations for the granular variant. The reason isn't more text or more keywords, it's clarity. A page that answers one question fully and directly is easier for an AI system to process than one that grazes ten questions at once.

That matches an observation that has held up since generative search emerged: breadth sells well to humans, because a comprehensive page feels competent. To a system reading individual passages, that same breadth reads as dilution.

Why granularity matters more than schema.org on its own

Granularity works at the level of the content chunks themselves, while schema.org only makes explicit what's already there structurally — which is why markup can't substitute for a clear page structure. In From SEO to GEO I already covered the mechanics behind generative search systems: an AI system splits a page into chunks, turns them into vectors via an embedding model, and compares them against the query. That's exactly where granular pages pull ahead.

A broad page with ten subtopics produces ten chunks that are all somewhat diluted — every paragraph carries a bit of the page's overall context, but is rarely the one sharp answer to a specific question. A narrow, focused page, by contrast, produces chunks that circle almost entirely around one topic. Vector similarity to the query is higher, the odds of being selected go up — and that's exactly what explains the 2.3× from the analysis.

Structured data following schema.org reinforces this effect, but doesn't replace it. It makes explicit what an AI system would otherwise have to guess — which entity a page is about, what attributes it has, how it relates to other entities. But markup alone won't rescue a page that blends ten topics into one wall of text. It's an amplifier for clarity, not a substitute for it.

Google itself clarified this recently: there's no dedicated schema format just for AI answers, and structured data isn't a requirement for generative search. What matters is that a page is unambiguous for humans and machines alike. Schema.org helps a great deal with that — it's a tool, not a fix for an unclear page.

What "granular" means in practice

Granularity doesn't mean shrinking every page to a single sentence. It means: one page, one clearly scoped topic, one clearly identifiable audience or question. In practice, that means:

• One question per section, answered fully. Don't mention something "also relevant" in passing — write each section so it can stand alone and be cited, without the AI needing to have read the paragraph before it.
• Specific pages instead of catch-all pages. One page per service, per audience, per use case — instead of one page trying to explain everything at once and, as a result, never giving the sharpest answer to anything.
• Name entities explicitly. Who, what, for whom — not left for the reader (or model) to infer from context, but stated directly in the sentence and backed up in the markup.
• Redundancy across pages isn't a problem. Unlike classic ranking, where duplicate content gets penalised, an AI system has no issue with two pages covering the same core point from different angles — as long as each page is clear on its own.

A practical example: a single "Services" page that works through ten offerings in ten paragraphs almost always loses out to ten short, dedicated pages, each with one offering, one audience and a clear answer to the most obvious question attached to it. To a human, the catch-all page might look tidier. To a system comparing chunks, it's ten diluted matches instead of ten sharp ones.

Headings are the real fault lines

Something that gets missed in the granularity discussion: most RAG pipelines don't split a page arbitrarily — they split it along its heading structure. An H2 section tends to become its own chunk, sometimes together with the H3 beneath it. That means your heading structure isn't just navigation for humans; it's the actual fault line an AI uses to cut your page into citable units.

Two practical consequences follow. First: a heading should name the question the section answers, not a vague umbrella term. "Pricing" is a weak heading for a chunk; "What does X cost per month" is a strong one. Second: the first sentence after a heading should carry the answer directly, not arrive after two sentences of context. If the core point lands in the third sentence, there's a real risk that exactly the chunk that could have been cited gets cut off mid-way or arrives incomplete in the model's context window.

Those two points — topically narrow pages and precise headings with the answer up front — are, together, the actual lever behind the 2.3× figure. Neither requires a new tool, just a different order when you write.

The thinking that slows most sites down

The common mistake: more content on a page feels more complete, and therefore higher quality. For humans that might hold, as long as the page is well structured. For a system that reads in chunks and cites individual passages, what matters isn't a page's overall length but how sharp the one passage in the running as an answer is. Whoever cuts their content so that every page fully answers one question is playing by exactly the rules the 2.3× from the analysis describes — regardless of how much schema markup ends up on the page.

It's not a one-off task you tick off either. New services, new audiences, new everyday questions keep shifting what still counts as sharp enough — a page that's granular enough today might, a year from now, be quietly bundling two topics that by then deserve separate answers. Treating this as an ongoing process rather than a finished project is what keeps you ahead.

For how structured data and entity clarity feed into a site's architecture in practice, see the AI discoverability (GEO) page.

Source

• AI Search Optimization Guide: LLM Visibility Strategies — ALM Corp, 2026. almcorp.com