Earlier this month we published a study of the citation corpus behind the August 2026 CiteHawk AI Index, and the headline was disagreement: the engines barely cite the same sources, and most of what they cite never appears on Google's first page.
This analysis asks the follow-up question that matters most if you own a brand: of the pages the engines do cite, who controls them?
The scope is the subset of that corpus where "own website" has a clean meaning: the 145 Index categories ranked at business level (rather than product model level). That is 10,178 citations pointing at 5,921 distinct pages across 2,743 websites, collected in the first week of August 2026 from ChatGPT, Claude, Grok, Perplexity, and Bing Copilot. Gemini appears in the earlier study but is excluded here: its stored citation rows carry a Google redirect host rather than the destination page, and own-site attribution needs a verified destination for every citation, so we left it out rather than guess. Google AI Overviews returned almost no citations in this corpus (3).
The attribution rule is strict and mechanical: a citation counts as "own site" when the cited domain equals, or is a subdomain of, a ranked brand's own domain in the same category. Everything else is third party.
Finding 1: Own websites supply 16% of citations. The other 84% is pages the business does not control.
Across all 10,178 citations, 1,624 pointed at the recommended businesses' own websites. That is 16.0%. The remaining 84.0% pointed somewhere else: a directory, a review platform, a publication, a community thread, a video, somebody else's blog.
Put plainly: when an AI engine recommends a business, roughly five out of six sources behind that recommendation are pages the business has no direct control over.
One number-hygiene note, because this figure will travel: this 84% is about page ownership. It is a different measurement from the 71% to 87% Google-overlap range in our earlier study, which was about whether cited domains appear in Google's top 10. Both happen to land in the eighties; they answer different questions.
Finding 2: The citing pages are directories and publications first, and the tail is enormous
We hand-classified the 120 most-cited websites, which carry 44.6% of all citations (4,538 instances). The mix:
- Directories, review and comparison sites: 41.8%
- News and publications: 25.7%
- Community and video (Reddit, YouTube, LinkedIn, Wikipedia): 16.3%
- Other companies' blogs and listicles: 10.0%
- Research, analyst, government and press-release pages: 6.2%
Classification is by what the site is, so a small slice of these citations (4.9%) are cases where the cited site is itself a ranked business in that category, a payments company cited in the payments category, for instance. The exact ownership split is Finding 1's per-category measure.
The tail deserves its own sentence. Those 10,178 citations spread across 2,743 distinct websites, and most of those sites were cited once or twice. The engines are not reading from one canonical list; they assemble answers from a long, messy tail of third-party pages.
Community reach is wide but engine-conditional: Reddit was cited at least once in 123 of the 145 categories and YouTube in 116, driven almost entirely by the engines with community appetite. Our earlier study measured that appetite per engine, and it spans two orders of magnitude, so a community strategy moves some engines and does approximately nothing on others.
Finding 3: In the categories closest to local business, own websites disappear entirely
The 16% own-site share is not evenly spread. At one end sit software and online-services categories, where vendors' own pages still hold ground: appointment scheduling software at 48.1%, expense management at 48.6%, time tracking at 40.8%. Online tutoring, at 54.2%, was the only category out of 145 where own websites supplied the majority of citations.
At the other end, own-site share collapses. In 49 of 145 categories, own websites supplied less than 10% of citations. In seven categories they supplied exactly zero: home insurance, home warranty companies, internet providers, mortgage lenders, pest control companies, tire brands, and antivirus software. Meal kit services sat at 1.1%.
Read that list again: home services, insurance, lenders, providers. The categories where own websites vanish from the AI reading list are precisely the ones local and consumer service businesses live in.
Two worked examples from the corpus. In pest control, the most-cited sources were This Old House, Money.com, Forbes, Bob Vila, BestCompany, the EPA, Today's Homeowner, and ConsumerAffairs. Not one ranked company's own website appears in the category's citation set. Home insurance is the same story with different mastheads: NerdWallet, J.D. Power, Money.com, U.S. News, ValuePenguin, The Zebra, Consumer Reports. The engines decided who the best insurers were without citing a single insurer's website.
For a business in a category like these, the AI answer about you is written on a handful of third-party pages, and most businesses have never audited what those pages say.
Finding 4: How much an engine trusts official websites varies six-fold
Own-site citation share per engine, on the same corpus:
- ChatGPT: 41.6% of its 1,525 citations pointed at recommended brands' own sites
- Grok: 13.7% (of 2,487)
- Claude: 13.7% (of 1,909)
- Perplexity: 10.2% (of 2,851)
- Bing Copilot: 6.9% (of 1,403)
ChatGPT leans on official websites six times as much as Bing Copilot does. A brand can be represented through its own pages in one engine's answers and assembled entirely from third-party sources everywhere else. This is the ownership version of the per-engine conditionality we keep finding in this corpus: testing one engine tells you about one editor's sources, not about your visibility.
What this means if you own a brand
- Audit the third-party layer before touching your website. In most categories, the pages that decide your AI visibility are directories, review platforms, and publications. Find out which ones your category's answers actually cite, then check whether you are present and accurately described on them.
- Let your category set the priority. A software vendor still wins real citation share through owned pages. A home services or insurance brand should assume near-zero own-site citation and put the work into the lists, reviews, and coverage the engines read instead.
- Treat third-party accuracy as a live risk. An outdated address on a directory, a discontinued service on a review page, a wrong price band in a comparison article: these are no longer citation-cleanup chores. They are raw material the engines quote as fact.
- Condition everything on the engine. Own-site reliance spans 7% to 42%, and community appetite spans two orders of magnitude. A tactic that moves one engine can be a no-op on another.
Every number above is computed from the August 2026 Index citation corpus and carries its sample size; the frozen editions publish their full answer corpora for independent recomputation. If you want to know which sources the engines cite in your category, and whether you are on that list, that is exactly what CiteHawk measures.
