Content
What AI actually cites — and what a cited page looks like. The page types models reach for versus the ones they only crawl, then the on-page signals the most-cited pages share. This is the “what should I publish?” layer.
Crawled vs cited
Crawlers visit everything; models cite only a little of it. Each page type’s share of citations against its share of crawls — and which way the trade falls.
What a cited page looks like
The most-cited pages compared with the rest of the field — bottom 50% to top 10%. Structured data and tables separate them; raw length barely does.
Schema that shows up on cited pages
How much more often each schema type appears on AI-cited pages than on the open web.
Lift is how many times more common a type is on cited pages. Person, ImageObject and Article mark the authored, well-described pages models trust — not the raw count of schema.
The FAQ effect
Average citations a page earns, by how it handles FAQs.
The schema alone isn’t the trick — it’s real Q&A content marked up so a model can lift the answer cleanly.
The most-cited pages
Real pages AI reaches for most, with the schema each one ships. Concrete proof of the patterns above.
| # | Page | Citations |
|---|---|---|
| 1 | 218 | |
| 2 | 174 | |
| 3 | 123 | |
| 4 | 111 | |
| 5 | 82 | |
| 6 | 80 | |
| 7 | 80 | |
| 8 | 79 | |
| 9 | 75 | |
| 10 | 75 | |
| 11 | 72 | |
| 12 | 72 |
The crawl-to-cite efficiency comes from matching 337K citations against 11M crawler visits and classifying each page by type. The blueprint and schema lifts come from 1.5K AI-cited pages across 950 domains, compared with open-web baselines. Efficiency is citation share ÷ crawl share; lift is how much more common a signal is on cited pages.