AI Assistants Cite Third-Party Lists, Not Your Homepage. A 60,000-Citation Study Shows How Little Your Own Site Counts.
A study of 60,350 AI citations found homepages earn just 4-7% of them. Third-party lists win 75-81%. What that means for brand visibility.
When ChatGPT and Claude answer a commercial question, they almost never cite the brand's own homepage. A July 2026 study of 60,350 citations found homepages earned just 7% of ChatGPT's and 4% of Claude's. Ranked lists and comparison pages took roughly 75-81%. In AI-first discovery, your site is not the surface that decides whether you appear.
What the data actually shows
The study, published by citations.press and reported in industry press, analyzed 60,350 citations generated across 10,000 commercial prompts spanning 50 industries, run through ChatGPT and Claude. The headline is stark: when these assistants answer a buying question, they overwhelmingly reach for third-party ranked lists, comparison pages, and category roundups — not the pages brands spend the most money building.
Two numbers carry the story. Homepage citations landed at 7% for ChatGPT and 4% for Claude. Third-party ranked and comparison pages captured roughly 81% for ChatGPT and 75% for Claude. The gap is not marginal. It is the difference between owning the surface and renting a mention on someone else's.
The two engines also disagree about who is authoritative. ChatGPT and Claude shared only about 18% of their cited domains. Claude drew on 208 unique domains against ChatGPT's 110. A brand cited confidently by one can be absent from the other — which means visibility is not a single score but a per-engine distribution.
The corroborating signal: ghost citations
The citations.press finding does not stand alone. In June 2026, Semrush — working with Kevin Indig and Growth Memo — published a parallel study of 3,981 domain appearances across 115 prompts in 14 countries. It found that 62% of AI citations never lead to a brand mention. The source gets linked; the brand name never makes it into the answer text a reader actually sees.
The engine-level detail is where it gets useful. Gemini mentioned brands 83.7% of the time but cited only 21.4%. ChatGPT reversed the pattern — citing 87% of the time but mentioning brands in just 20.7% of answers. Being cited and being named are two different wins, and no engine gives you both by default.
Put the two studies together and the picture sharpens: your homepage is rarely the source, and even when your content is the source, the model often paraphrases without crediting you. Both failure modes route around the pages you control.
By the numbers
| Signal | What the data shows | Source |
|---|---|---|
| Homepage share of citations | 7% (ChatGPT), 4% (Claude) | citations.press, July 2026 |
| Third-party list / comparison share | ~81% (ChatGPT), ~75% (Claude) | citations.press, July 2026 |
| Cited-domain overlap between engines | ~18% | citations.press, July 2026 |
| AI citations that never mention the brand | 62% | Semrush + Growth Memo, June 2026 |
Why optimizing your own site misreads the shift
The instinct is to respond by rebuilding the homepage — more schema, cleaner copy, tighter product pages. That treats this as an on-page problem. The data says it is a source-selection problem. A model assembling an answer pulls from the handful of pages it already trusts for that query. If none of them is yours, no amount of homepage polish inserts you into the answer.
This is the distinction Jaxon Parrott drew when he coined the term Machine Relations: being findable by a person scanning results and being usable by a model assembling an answer are two different disciplines. The citation data is Machine Relations made measurable. The model is not ranking your page. It is deciding which sources to synthesize, and it heavily favors independent third-party ones over anything self-published.
That preference is rational from the model's side. A "best CRM for startups" listicle on a trusted outlet reads as an assessment; a brand's own homepage reads as a claim. The assistant, optimizing for a trustworthy answer, weights the assessment. That is earned authority functioning exactly as the category predicts — trust accrues to sources the model treats as independent, and it is the primary driver of AI visibility.
What the winning move looks like
The brands holding their discovery are not the ones with the most polished homepage. They are the ones present in the third-party sources the model already cites — the ranked lists, the category comparisons, the roundups in outlets an assistant treats as neutral. Getting there is placement and earned mention, not on-site optimization.
The market is starting to price this in. Adobe launched Adobe Brand Visibility in June 2026 as a dedicated product for tracking how brands surface inside AI answers — a signal that measurement of this layer is becoming table stakes, not a fringe experiment.
The practical sequence is straightforward, if not easy:
- Map the sources, not the keywords. For each priority query, identify which third-party pages the assistants actually cite. That set — not the SERP — is your real competitive field.
- Earn placement in that set. Get accurately represented in the comparison pages, category lists, and trusted outlets the model reaches for. This is earned media discipline, not SEO.
- Measure per engine. Because ChatGPT and Claude overlap on only ~18% of domains, a single visibility number hides more than it reveals. Track presence engine by engine. Tools like AuthorityTech's publication intelligence map which publications AI engines cite most, so placement effort targets the sources that actually move answers.
The uncomfortable implication for most 2026 budgets: the money is pointed at the homepage, and the homepage is where AI assistants look least.
FAQ
If my homepage barely gets cited, should I stop investing in it? No — the homepage still converts the traffic you earn and anchors your brand identity. But it is not the lever for AI visibility. Treat it as a conversion asset, and treat third-party placement as the discovery asset. They are different jobs with different playbooks.
How do I find which third-party pages the AI actually cites for my category? Run your priority commercial prompts through ChatGPT, Claude, and Gemini and record the cited sources, or use a tracking tool that captures citations at scale. Because engines share only about 18% of their cited domains, build the source map per engine rather than assuming one list covers all of them.
If AI assistants are answering questions about your category without citing you, the first step is knowing where you stand. A visibility audit shows which engines cite you, which cite competitors, and which third-party sources are shaping the answer.