Which pages on a brand's own site earn AI citations?

An owned citation is a citation in an AI answer that points to a page on the brand's own website. We classified every cited address by subdomain and path pattern into ten page types. Blog and editorial pages supplied 1,578,724 of 5,635,394 owned citations, or 28.0%. Comparison and alternatives pages supplied another 1,048,206, or 18.6%. Together that is 2,626,930 citations, or 46.6% — nearly half.

A widely shared small study of 240 pages on four sites reported that product-style pages beat blog posts 76% to 24%. Across 236,813 brand websites we observed the reverse: product and commerce pages carried 6.5% of owned citations, about one seventh of the blog-plus-comparison share. When AI engines cite a brand's own site, they mostly cite the pages that explain and compare, not the pages that sell.

Unclassified addresses were the largest single bucket at 32.2%; the last two insights cover them. Among the nine named page types, blog and editorial led at 28.0%, followed by comparison and alternatives at 18.6%. The homepage came fourth overall at 10.2% (575,497 citations), ahead of product and commerce at 6.5% and docs and help at 3.0%. Community and forum pages, company pages, pricing pages, and customer-story pages each carried under 0.6%.

Our companion study of homepage vs deeper links found two thirds of owned citations go below the homepage; this mix shows where they land. The homepage still out-earned pricing pages about 59 to 1.

Within the 1,048,206 comparison and alternatives citations, pages with best-of or top-N addresses carried 838,227, or 80.0%. Head-to-head pages (vs and compare addresses) carried 14.7%, and alternatives pages 5.4%.

Best-of pages alone carried 14.9% of ALL owned citations — one in seven. Our page-level study of all cited domains found six in ten of AI's most-cited pages are best-of listicles; this shows the same format working when the list lives on a brand's own site.

Pricing pages carried 9,835 owned citations, or 0.17% — about 1 in 573. Customer-story pages carried 4,971, or 0.09%, and company pages (about, careers, press) 29,511, or 0.52%. Only 1,815 brand sites earned a pricing-page citation at all.

These are the pages sales teams polish hardest, and AI engines almost never cite them. This describes where citations landed, not why; pages behind menus and pages with little quotable text give an engine little to cite.

Averages can hide brand-level variety, so we also asked, for each of the 16,956 brand sites with at least 50 owned citations, which page type carried the most. Blog and editorial led for 38.5% of them. Unclassified addresses led for 32.6%, comparison pages for 14.3%, and the homepage for just 7.7%. When a page type leads a brand's mix, it carries about two thirds of that brand's owned citations on average.

Named examples: Guideflow's blog carried 84.8% of its 5,785 owned citations, Wiz's 82.7% of 2,075, and Sprout Social's 81.6% of 2,206.

We bucketed each brand by its average rank in its top market over the window. Brands averaging rank 1 to 3 drew 17.4% of their owned citations on the homepage; brands ranked 11 or lower drew 8.4%. Comparison pages inverted: 9.3% for top-3 brands versus 19.7% for rank-11-or-lower brands — 2.1 times as high.

AI engines cite a market leader's site for what the brand is, and a smaller brand's site for content that ranks options. For a brand outside the top of its market, comparison and best-of content is the page type earning citations right now.

The mix shifts by industry. Information Technology brands drew 9.4% of owned citations on docs and help pages; Sales and Marketing brands drew 0.9%. Software brands had the highest comparison share at 23.4%; Sales and Marketing brands had the highest homepage share at 15.7%. Blog and editorial stayed between 22.8% and 35.8% everywhere — the one page type that carries citations in every industry.

Compare your own mix with your industry row, not with the overall average.

Docs and help pages carried 170,564 owned citations, or 3.0%. Vendor advice often calls documentation the most underrated GEO asset; the base rate says docs are a small slice of owned citations for most brands. The exception is developer-facing brands, where docs ARE the cited site.

Microsoft Learn drew 100% of its 6,948 owned citations on docs pages, as did Shopify Help Center, GitHub's docs site, Atlassian Support, and Square Developer. Android drew 96.7%, PyTorch 74.7%, NVIDIA 74.1%, HashiCorp 69.5%, and OpenAI Platform 64.7%. The industry table above shows the same split: 9.4% docs share in Information Technology, 0.9% in Sales and Marketing.

Before exclusions the raw corpus held 7,018,229 owned citations. We removed 1,382,835 of them, or 19.7% — links pointing at 33 platform sites (search engines, social networks, UGC hosts, and app stores) plus Parse's own site. On the platform sites the cited pages are overwhelmingly not the company's own publishing: 99.9% of YouTube's owned citations pointed to video and user paths, 99.4% of Reddit's to subreddit threads, and 94.4% of Google's to search and redirect addresses.

Counting a Reddit thread as Reddit-the-brand's owned page would corrupt every mix figure above, so the study corpus removes these sites. The table shows the largest.

1,817,070 owned citations, or 32.2%, pointed to addresses that match none of our nine named page types. 40.5% of those are single-segment addresses — a slug directly under the domain root, like a landing page or a one-off tool.

The unclassified share is a real finding, not just noise: a large part of what AI engines cite on brand sites does not follow a standard site template. Reading the mix as shares of classified links only would overstate every named type by roughly half, so we report shares of ALL owned citations and show the remainder honestly.

We summed the owned-citation daily layer of the public Parse index from May 24 through August 16, 2026 — a window in which the index tracked ChatGPT Search and Google AI Mode; the layer combines both engines without a split. Each cited address was stripped of query parameters and fragments, lowercased, cleared of up to two leading locale segments, and matched against subdomain and path patterns in a fixed order (first match wins). An independent aggregation lane at the prompt grain reproduced the corpus total of 5,635,394 exactly, and a page-grain recount (distinct pages instead of links) kept the same top three types; homepage and product swap places at that grain, because a homepage is one page cited many times. Besides the platform sites above, we excluded Parse's own site, so the study never measures itself.

Limits: the pattern classifier covers 67.8% of links; page types describe addresses, not page content; the layer has no engine or answer-position dimension; brand rank buckets use the brand's average rank in its top market over the window. Counts describe where citations landed in this window, not why.

Blog and comparison pages carried 46.6% of owned citations; pricing pages carried 0.17%. Best-of lists were 80% of comparison-page citations. Top-3 brands drew homepage citations at twice the rate of lower-ranked brands, and lower-ranked brands drew comparison citations at twice the rate of leaders.

If you want AI engines to cite your site: publish guides and ranked lists on your own domain for the questions your buyers ask, and put the facts you want quoted — including prices — inside citable editorial pages, because the pricing page itself will almost never be the citation. If you sell to developers, treat docs as the primary citation surface. Then rerun the same page-type audit on your own market each quarter.