The Anatomy of a Page That AI Cites

Sill's monitoring pipeline queries actual chat interfaces daily across ChatGPT, Perplexity, Gemini, and Google AI Overviews. For every page cited in an AI response, we record provenance metadata: which platforms cited it, how frequently, in response to which queries, and across which product subcategories.

The foundational GEO paper (Aggarwal et al., KDD 2024) established the principle that content-level optimizations affect AI citation rates, finding 30-40% improvements from adding statistics. The question our data answers is practical: across 22,785 real-world pages being cited by real AI platforms, what specific patterns emerge?

Across 22,785 cited pages, ChatGPT is the source of 60.6% of all page-level citations. Perplexity and Google AI Overviews each account for roughly 22%, while Gemini trails at 4.8%. This distribution reflects platform market share (ChatGPT holds roughly 79% of global generative AI web traffic, per Similarweb's GenAI Index), but the absolute numbers mask the more important finding.

The important finding is not which platform cites the most pages. It is which pages get cited by multiple platforms. That is where the real quality signal lives.

Platform diversity is the number of distinct AI engines that independently cite a given page. In our dataset of 22,785 pages, 91.5% are cited by exactly one platform. Only 1,945 pages (8.5%) achieve citation from two or more platforms. Just 277 pages (1.2%) reach three or more. And only 27 pages out of 22,785 (0.1%) are cited by all four platforms we track.

This matters because each AI platform has different citation DNA. ChatGPT relies on Bing's index and favors Wikipedia. Perplexity runs real-time web searches and favors Reddit. Google AI Overviews pull from organic top-10 results. Gemini favors authoritative lists and local reviews. When a page crosses all of these different retrieval systems, it has demonstrated a quality that transcends any single platform's bias.

The most common multi-platform combination is ChatGPT + Google AI Overviews + Perplexity (28 pages), followed by the full four-platform set of ChatGPT + Gemini + Google AI Overviews + Perplexity (13 pages). These pages are worth studying because they represent the content that passes every major AI retrieval system's quality filter independently.

If your goal is to understand what makes a page AI-citable, platform diversity is the metric to optimize for. A page cited by four platforms is more informative than a page cited ten times by one platform, because the former has demonstrated universal relevance.

The citation frequency distribution follows a steep power law. Out of 22,785 pages, 20,315 (89.2%) are cited exactly once. Only 2,470 pages (10.8%) are cited two or more times. Just 579 pages (2.5%) reach three or more citations. At the far end, only 10 pages in the entire dataset achieved 11 or more citations.

This distribution has a direct strategic implication. Most pages that get cited by AI are one-time retrievals for a specific query. The pages that get cited repeatedly are the ones that answer multiple related queries. Breadth of relevance within a topic area is what drives repeat citation, and our subcategory data confirms this.

For each cited page, Sill tracks which product subcategories triggered the citation. A page cited for "Contact Center as a Service" only has a subcategory count of 1. A page cited for both "RV Maintenance & Repair Services" and "RV Parts & Accessories" and "Recreational Vehicles" has a subcategory count of 3. The relationship between subcategory coverage and citation frequency is the strongest signal in our data.

Pages relevant to 3+ subcategories average 4.5 citations, compared to 1.1 for single-subcategory pages. That is a 4.1x multiplier. The mechanism is straightforward: a comprehensive comparison page that covers multiple product angles provides relevant answer fragments for a wider range of buyer queries. AI retrieval systems retrieve it for "best X for Y," "X vs Z," and "how to choose X" all from the same page.

This is the content strategy takeaway. Narrow pages that answer a single query get cited once and forgotten. Comprehensive pages that span adjacent subcategories become citation magnets. The data shows a near-linear relationship: each additional subcategory a page is relevant to adds roughly one additional citation.

Across our full dataset, Reddit leads in raw page count with 408 cited pages. YouTube ranks fifth by page count (104 pages) but leads in a more important metric: citations per page. YouTube pages average 3.4 citations each, compared to 1.0-1.1 for most other domains. A single YouTube video review generates more total citations than three average Reddit threads.

YouTube's 3.4x citation efficiency stands out. A YouTube video review is cited across multiple queries and often by multiple platforms (YouTube dominates the multi-platform citation list, with 26 of the 100 highest-diversity pages). The Ahrefs study reported a 0.737 correlation between YouTube presence and AI visibility. Our data confirms this: YouTube pages are cited more frequently, across more platforms, than pages from any other domain.

The data also shows that specialized review sites (rtings.com at 1.6, tomshardware.com at 1.9, pcgamer.com at 1.5) outperform general-purpose platforms in citations per page. Depth of expertise within a category correlates with citation efficiency. Chen et al. (University of Toronto, 2025) found that 69-82% of AI search citations come from earned media. Our domain distribution confirms this: the top cited domains are overwhelmingly third-party sites, not brand-owned properties.

The absence of certain traditional SEO signals in highly cited pages is as informative as the presence of the traits above.

The SearchAtlas study of 21,767 domains measured domain authority correlation with AI visibility at r = -0.12 for ChatGPT and -0.18 for Perplexity. These are not weak correlations. They are slightly negative. Our data is consistent: the top cited domains are a mix of high-authority (Wikipedia, Forbes, Gartner) and moderate-authority (rtings.com, soundguys.com, techbloat.com) sites. Domain authority does not predict which pages get cited. Content relevance and structure do.

Synthesizing the patterns across our dataset, the pages with the highest citation frequency and platform diversity share a consistent set of traits. These compound: a page with all of these traits is cited more frequently than a page with any subset.

Knowing what makes a page citable is useful. Knowing which of your pages are actually being cited, by which platforms, in response to which queries, is actionable.

Sill's monitoring pipeline tracks citation provenance at the page level. For every page cited in an AI response about your brand, we record the citing platforms, the queries that triggered the citation, the citation frequency over time, and the subcategory associations. Across 22,785 pages and 11,405 domains, this creates a dataset that reveals not just whether your brand appears, but which specific pages are doing the work.

The three-layer attribution model we described previously starts with this data. Simulated visibility (daily SOV tracking) tells you whether your overall presence is growing. Citation provenance data tells you which specific pages are driving that growth and which need attention.

The data in this post demonstrates the scale of the challenge. With 89% of pages cited only once and only 8.5% achieving multi-platform visibility, the difference between a page that contributes to AI presence and a page that does not comes down to the structural and topical traits described above. Without page-level citation data, content optimization is directionless. With it, every content investment can be targeted at the specific gaps that matter most.