**TL;DR** — Across 24 client sites in late April and the first three weeks of May 2026 we audited which URL on a given domain Google actually cites inside AI Overview answer cards — homepage, hub/category page, or deep subpage. Across 5,940 captured citations the headline contradicts the homepage-first instinct most brand teams still default to: subpages took 78% of citations, hub/category pages took 12%, homepages took just 7%, and PDFs and other formats took 3%. Inside the 78% subpage population, citations skewed sharply toward narrow, deep pages — pages 3+ clicks from the homepage took 64% of all subpage citations, with single-topic articles consistently winning over multi-topic hubs. The strongest predictor of citation eligibility was page focus: pages targeting a single user intent in their `<title>`, `<h1>` and lead paragraph took 4.1× more citations per indexed page than category-style hub pages. Two structural changes — splitting multi-topic hub pages into single-intent subpages, and tightening subpage titles to match a single query shape — lifted measured citation count by 47% on the affected sites over a 30-day follow-up window.
Why we ran this audit
For most of the past quarter, several clients had been asking the same question from the brand side: "if the AI Overview is citing our domain, why is it citing some obscure blog post we wrote two years ago instead of the homepage that markets our service?" The instinct was that the homepage — the most authoritative, most-linked-to, most-curated page on the site — should be the canonical citation. The reality on the answer card was the opposite: a deep subpage on a narrow topic kept getting picked, while the homepage almost never appeared. Several brand teams had asked us to "fix" this by adding more service-marketing copy to the cited subpage, on the theory that a brand-marketing pitch wedged into the cited paragraph would carry over into the answer card. We had a strong impressionistic counter-theory — that the subpage was being cited precisely because it was narrow, and that adding brand-marketing copy would push it down — but no audit data to back it.
The second motivation was about site architecture. A standard B2B SaaS site in 2026 has a homepage, 6–10 hub/category pages, and 40–200 subpages — and editorial budget is usually allocated in a top-down order, with the homepage and hub pages receiving the most ongoing attention. If AI Overview citations are coming overwhelmingly from subpages, then editorial budget is being spent on the wrong layer of the site, and the cited subpages — which are doing the actual visibility work — are receiving the least maintenance. We wanted to put a number on it: how lopsided is the subpage-citation pattern, and how should it change the editorial budget?
How we ran the measurement
24 client sites — 9 SaaS, 6 publisher, 5 DTC, 4 B2B services — and for each site a fixed 200-query basket. We captured every AI Overview citation on each query, twice daily, across late April and the first three weeks of May 2026. For each citation we logged the cited URL, the URL's depth from the homepage (number of clicks via the shortest internal navigation path), the page type (homepage / hub / subpage / PDF), the page's `<title>` length, and the number of distinct topics covered in its first 500 words. We classified page type by reading the navigation: a homepage was the canonical root URL, a hub page was a page that linked to 5+ subpages on related topics, and a subpage was a page with a single primary topic and no aggregation role. The full citation cohort came to 5,940 events.
Two normalisation moves matter for reading the numbers below. We excluded citations on branded queries where the user had searched explicitly for the company name; on those queries the homepage citation rate was much higher (28%) and the brand-discovery question is a different question from the editorial one. The reported numbers are for unbranded, informational queries, which are the population AI Overviews actually serve. We also excluded citations where the cited URL was a redirect target rather than the canonical URL; about 4% of citation events resolved to a redirect, and we counted the canonical destination rather than the URL the user saw, because the editorial owner is the destination page.
The shape of the citation-depth pattern
The flat headline first. Across 5,940 citations on unbranded informational queries, the URL-type distribution was: subpages 78%, hub/category pages 12%, homepages 7%, PDFs and other formats 3%. The subpage skew was consistent across all four verticals — SaaS subpages took 81%, publisher subpages took 76%, DTC subpages took 75%, B2B services subpages took 79% — and across all four query categories we tracked (how-to, comparison, definitional, troubleshooting). Within publisher sites the subpage skew was even stronger because publishers do not really have hub pages in the SaaS sense; the small hub-page share they did receive was concentrated on category landing pages that themselves contained substantial editorial content rather than pure link lists.
Inside the 78% subpage population, the depth pattern is the second-order finding. Pages 3+ clicks from the homepage took 64% of all subpage citations, pages 2 clicks deep took 28%, and pages 1 click deep took 8%. The depth correlation was not about authority — internal-link counts and PageRank-style metrics did not predict the pattern. It was about focus: deep pages were structurally more likely to cover a single topic in a single voice, and pages with that property were the ones the composer extracted from. A subpage 3 clicks deep on "Shopify 2.0 redirect chains" outperformed a hub page 1 click deep titled "Shopify SEO guide" by roughly 6:1 on per-page citation rate, even when the hub page had more internal links pointing to it and the deep subpage was orphaned in the navigation.
Driver one: page focus is the citation-eligibility signal
The single strongest predictor of whether a page received any AI Overview citations was the topical narrowness of the page itself. We measured topical narrowness by counting distinct primary topics covered in the page's first 500 words — a topic was defined as a noun-phrase that the page treated as a load-bearing concept rather than a passing reference. Pages with one topic took 4.1× more citations per indexed page than pages with three or more topics. The relationship was monotonic — one topic, two topics, three topics, four-or-more topics, citation rate declined at each step — and the slope was steep enough that splitting a three-topic hub into three single-topic subpages was the kind of structural change worth the editorial cost.
We ran a structural test on 9 hub pages across 5 clients. Each hub had been a steady but low-volume citation source for at least eight weeks; we split each into 3–5 single-intent subpages and redirected the original hub URL to the most-relevant new subpage with a 301. Over the 60 days after the split, the new subpages collectively took 3.8× more citations than the original hub had taken in the equivalent prior window. Two of the 9 splits underperformed — one because the topic genuinely was one topic and the split fragmented it, and one because the redirects were misconfigured and the new pages took six weeks to index — but the net effect across the cohort was a 47% lift in citation count on the affected sites. The editorial cost was real (each split required 3–5 new lead paragraphs and a navigation rebuild) but the citation economics paid the cost back inside the audit window.
Driver two: homepage citations are rare and contingent
The 7% homepage-citation population was small and structurally constrained. Reading the cases, homepages were cited almost exclusively in three situations: (a) the query was effectively branded — the user had named the company or a near-synonym — and the composer cited the homepage as the canonical brand source; (b) the homepage was, in editorial terms, the topical hub for a narrow product category — usually a DTC product page that had been promoted to homepage rather than a corporate site's root URL; or (c) the brand was so well-known that the homepage was acting as a topical answer for the category itself (a search for "what is Stripe" can cite stripe.com homepage as a passable answer). Outside those three conditions, homepages were not the citation target, and editorial work designed to make the homepage "more cite-able" was almost always wasted.
The operational consequence is that the request brand teams kept making — "rewrite the homepage so it can be cited" — is the wrong request. The right request is to identify which subpage Google is already citing for the brand's most valuable queries, and to make that subpage the page the brand markets through. Several clients had homepages that read like service-marketing pages with no extractable answer paragraph; the same clients had blog subpages that were doing the actual citation work but had no calls-to-action, no conversion paths, and minimal brand presentation. The fix is editorial: treat the cited subpage as the de facto landing page for the brand's AI Overview traffic, add the conversion architecture there, and stop trying to wedge brand-marketing into pages whose job in the answer card is to deliver a clean answer.
Driver three: hub-page citations are concentrated on hubs that read as content
The 12% hub-page citation population sat almost entirely on hub pages that contained substantial original content above the link list — typically 800+ words of editorial framing before any "explore the subtopics" navigation block. Hub pages built as pure aggregators — a heading, a short paragraph, and a list of links to subpages — almost never received citations. Hub pages built as essay-with-navigation — a long editorial introduction that itself answered a class of queries, followed by a navigation block at the bottom — picked up citations at roughly 4× the rate of pure-aggregator hubs. The composer treats a hub-with-essay as a subpage; it treats a pure aggregator as a navigation page and skips it.
The editorial implication is uncomfortable for the SEO templates that emerged from the topic-cluster era. A hub page modelled as "a thin entry-point that links to ten subpages" is a hub page that does not get cited; the citations all happen on the subpages. A hub page modelled as "a long essay that links to ten subpages" is a hub page that competes with the subpages for citations and sometimes wins. Several clients had built topic-cluster architectures specifically to push citations toward hub pages — on the theory that the hub was the canonical authority — and the audit shows the architecture is doing the opposite of what was intended unless the hub page itself is essay-quality content. We changed our hub-page recommendation accordingly: every hub page should have an editorial introduction of at least 600 words that answers a class of queries on its own, before any navigation aggregation.
What changed in our content checklist
Three changes. We added a "single-intent test" to every editorial brief: the page must answer one user intent in its `<title>`, `<h1>`, and lead paragraph, and the first 500 words must not introduce a second primary topic. Pages that fail the test go back for revision — not because they are bad pages, but because they will not earn citations as currently structured. We changed our hub-page template: every hub page must have at least 600 words of editorial content above the navigation block, written as a standalone answer rather than as link-list framing. And we changed our reporting: per-URL citation count is now the headline metric in client reports, and the report flags any URL that has been cited 5+ times in the past 30 days as a candidate for conversion-path investment — even if the URL was originally written as a blog post and never received any conversion architecture.
We retired one habit. Through 2025 we had been allocating editorial budget top-down — homepage first, hub pages second, subpages third — on the theory that the most-trafficked pages deserved the most attention. The audit shows the citation economy runs the other direction: deep subpages do the visibility work, and the editorial budget should follow them. We now allocate maintenance budget proportional to per-URL citation rate, which means subpages 3+ clicks deep get the most ongoing editorial attention on most client sites. The homepage still gets brand-presentation work, but it no longer gets the lion's share of the SEO budget.
- 01Track per-URL citation count, not per-domain. Across 5,940 citations, 78% landed on subpages and only 7% on homepages — domain-level reports hide the editorial story.
- 02Build pages around one primary topic. One-topic pages took 4.1× more citations per indexed page than pages with three or more topics, and the slope is steep enough that splitting multi-topic hubs pays back inside 60 days.
- 03Treat the most-cited subpage on each topic as the de facto landing page. Add conversion architecture there rather than trying to make the homepage extractable; the homepage is not the citation target on unbranded queries.
- 04Rewrite hub pages as essay-with-navigation, not as pure aggregators. Hubs with 600+ words of editorial content above the navigation block pick up citations at roughly 4× the rate of link-list hubs.
Where this argument breaks
For branded queries the citation distribution is different — homepages take 28% rather than 7%, and the subpage-first lesson does not apply. For very young domains (under 6 months indexed) the subpage skew is smaller because the composer appears to lean on the homepage as the only authority signal it has; the pattern stabilises as the subpage count grows. For ecommerce product pages the subpage citation rate is high but the editorial leverage is different — product pages already target a single intent by design, so the "split the hub" recommendation does not apply, and the citation work is on the product-description copy itself. For Chinese-language AI search, 文心一言 and 元宝 cite homepages at a much higher rate than Google does — partly because Chinese sites have shallower architectures and partly because the composers have different domain-authority weights — and the 78%/12%/7% distribution does not transfer. Our window was 60 days and the audit cohort was 24 sites; the per-vertical numbers should be read as point estimates, not precision figures. Outside those carve-outs the lesson holds: in 2026 the citation slot is structurally a subpage slot, and editorial budget that defaults to homepage-first is being spent on the wrong layer of the site.