Anatomy of one cited passage
An AI citation has a predictable anatomy. When ChatGPT or an AI Overview quotes a page, the passage it lifts tends to sit in the same place, read the same way, and come from the same kinds of sources.
Once you can see that pattern, getting cited stops being luck.
For years the goal was a ranking. Now the goal is a quote: a sentence or two that an AI engine extracts from your page and folds into its answer, sometimes with a link, often without one. The useful news is that recent large-scale studies have dissected thousands of real AI citations and found they share a structure. This article takes that citation apart piece by piece: where on the page it comes from, what makes a passage quotable, which sources engines favor, and why the click so rarely follows.
What is an AI citation?
An AI citation is the specific passage of source text that a generative engine extracts and uses to build part of its answer: whether or not it shows a visible link back to you. It’s narrower than a ranking and narrower than a brand mention: the engine isn’t endorsing your whole page, it’s quoting one extractable chunk of it that answered the question well.
That distinction matters because it changes the unit of optimization. Classic SEO optimized a page to rank. AI visibility optimizes passages to be quotable. A page can rank well and still never get cited if nothing on it is structured to be lifted cleanly. So the practical question isn’t “how do I rank this page?”: it’s “which two sentences on this page is an engine most likely to quote, and have I made them easy to find and trust?”
Where on a page do AI engines pull citations from?
Overwhelmingly from the top. In an analysis of 3 million ChatGPT responses and 30 million citations, narrowed to 18,012 verified citations, Kevin Indig found that 44.2% of citations come from the first 30% of a page’s content, with citation likelihood dropping steadily the deeper you go.
There’s a twist one level down. Within a paragraph, the model reads more deeply than you’d expect: 53% of cited text came from the middle of paragraphs, versus 24.5% from first sentences and 22.5% from last. The lesson isn’t to cram every answer into an opening sentence: it’s to front-load your key sections near the top of the page, then keep each paragraph dense with substance the whole way through rather than burying the payoff in a conclusion.
Traditional search rewarded depth and a delayed payoff. AI rewards immediate classification — clear entities and direct answers, surfaced early.
What makes a passage “quotable” to an AI?
Quotable passages share a recognizable shape: they define things plainly, name specifics, answer a question, and strike a measured tone. Indig’s analysis isolated five linguistic traits that separated heavily cited text from the rest, and none of them are keyword tricks.
- Definitive language. Cited passages were nearly twice as likely to use clear definitions — “X is…,” “X refers to…” Direct subject-verb-object statements beat vague framing.
- A question-and-answer structure. Cited content was 2× more likely to contain a question mark, and 78.4% of citations tied to questions came from headings. Engines often treat an H2 as the prompt and the paragraph beneath it as the answer.
- Entity richness. Typical English runs 5–8% proper nouns; heavily cited text averaged 20.6%. Named brands, tools, people, and numbers anchor an answer and reduce ambiguity.
- Balanced sentiment. Cited text clustered around a subjectivity score of 0.47 — neither dry fact-dump nor opinion piece. The winning tone reads like analyst commentary: fact plus interpretation.
- Business-grade clarity. Top passages averaged a Flesch-Kincaid grade level of 16, versus 19.1 for lower-performing text. Shorter sentences and plain structure beat dense academic prose.
Put plainly: the content engines quote is content that already reads like a well-sourced answer. That’s the same principle behind writing citable content: lead with a self-contained answer, name specifics, and back claims with numbers.
Which sources do AI engines cite most?
It depends heavily on the engine. There is no single “most-cited” list that holds across platforms. In Profound’s analysis of 680 million citations (August 2024 to June 2025), ChatGPT leaned on encyclopedic and editorial sources, with Wikipedia its single most-cited domain at 7.8% of all citations. Google’s AI Overviews and Perplexity skewed toward community and discussion content, with Reddit their top source (2.2% and 6.6% of citations respectively).
One pattern does hold: commercial sites dominate the destination. Across ChatGPT citations, 80.4% pointed to.com domains and another 11.3% to.org, so being a credible commercial publisher is table stakes, not a disadvantage. But which specific domains get favored is unstable. Semrush, tracking more than 100 million citations across 230,000 prompts over 13 weeks in 2025, watched ChatGPT’s Reddit citations swing from roughly 60% of responses in early August to about 10% by mid-September, and Wikipedia drop from about 55% to under 20%, while those same sources held steady on other engines.
An honest caveat: the position and language findings above come from a ChatGPT-focused study, and the source-mix data is a snapshot of a moving target. Treat them as strong signals about how extraction-based AI behaves, not fixed rules that apply identically to every engine on every day. The traits that travel best across platforms are the durable ones: clarity, specificity, and structure.
Why doesn’t an AI citation send much traffic?
Because the answer is delivered on the results page, most users never click through. Pew Research Center found in 2025 that when an AI summary appears, users click a traditional link just 8% of the time, versus 15% without one, and only about 1% click the link inside the AI summary itself. The citation is doing its job in the answer, not in your analytics.
That reframes why anatomy matters. The value of an AI citation is increasingly the mention itself (your brand named and described accurately at the moment a buyer is forming an opinion) rather than the session it sends. It also explains why standard reporting understates AI’s impact: the influence shows up later as branded search and direct visits, not as a tidy referral. Measuring it is the job of an AI visibility audit.
How do you engineer content to get cited?
You build the anatomy of a citation into the page on purpose. Working from the data above, four moves do most of the work:
- Front-load the substance. Put your clearest, most quotable answers in the top third of the page, where 44% of citations are drawn from — not saved for a grand conclusion.
- Write definitions and answers, not throat-clearing. Open sections with “X is…” statements and direct answers. Use H2s as the questions and the next paragraph as the answer.
- Pack in specifics. Name tools, brands, people, and numbers. Higher entity density is one of the strongest markers of cited text, and it’s the easiest to add.
- Earn broad, accurate brand presence. Engines quote sources they’ve seen described consistently across the web. That wider footprint is the heart of AI search and LLM visibility work, and it’s what separates a page that ranks from a brand that gets recommended.
If you want the upstream mechanics (how engines assemble that candidate pool in the first place) we break it down in how LLMs choose their sources and in the practical checklist for getting cited by ChatGPT.
An AI citation has a consistent anatomy: it’s pulled mostly from the top third of a page, it reads as a clear, entity-rich, plainly-worded answer in a measured tone, and it comes from a source the engine trusts, though which sources those are shifts by platform and over time. Getting cited is less about ranking harder and more about engineering quotable passages and a brand that’s described accurately across the web.
Is AI quoting you — or your competitor?
We’ll run your brand against real buyer prompts across ChatGPT, Gemini, Perplexity, and Google’s AI Overviews and show you exactly where the citation gaps are. Free, no obligation.
Sources
- Kevin Indig, “The science of how AI pays attention,” 2026: 3M ChatGPT responses, 30M citations, 18,012 verified; page-position and linguistic-trait findings (reported by Search Engine Land, Feb 2026)
- Profound, “AI Platform Citation Patterns,” 2025: 680M citations (Aug 2024–June 2025); ChatGPT/AI Overviews/Perplexity source mix and TLD distribution
- Semrush, “The Most-Cited Domains in AI: A 3-Month Study,” 2025: 100M+ citations across 230,000 prompts, 13 weeks; ChatGPT Reddit/Wikipedia volatility
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears,” 2025: 8% vs 15% click rate; ~1% click the in-summary link