TJ Digital analyzed 193,507 AI answers from ChatGPT, Google AI Mode and AI Overviews, drawn from 4,167 questions across 78 client companies and nine business models. The capture ran in weekly snapshots from August 27 to October 21, 2026. The question behind it: what separates pages generative engines actually cite from pages they skip?
A direct answer up front separates cited pages
Blind raters scored 195 pairs of pages without knowing which side had been cited. Pages cited by AI opened with a direct answer 20% of the time, versus just 5% for passed-over pages that ranked in Google’s top ten but never appeared in AI answers.
An earlier round told a similar story with a smaller gap. When the same AI answer brought up pages but cited only some of them, cited pages opened with a direct answer 23% of the time, compared with 14% for uncited pages. Being retrieved by AI is not enough; the pages AI actually cites are more likely to answer immediately.
What else was linked to a citation
After accounting for domain rating and page type, TJ Digital reported these odds of being cited:
- Year in the title: more than double the odds
- Publish or update date in the past 12 months: 88% higher odds
- Tables: 69% higher odds
- First-hand experience: 45% higher odds
- Prices: 28% higher odds
- Schema markup: 13% higher odds
Author bylines showed 28% lower odds, and calls to action showed 12% lower odds. Robertson says he cannot explain those two results and would not remove either element based on this alone. Completeness also differed: cited pages scored 2.3 on a 1-to-5 scale, while passed-over pages scored 1.7. Promotional tone was nearly identical.
Why it matters and what to test
Google Top 10 visibility no longer guarantees generative visibility. Half of the passed-over pages were missing an answer anywhere in the portion raters reviewed, versus 23% of cited pages.
But Robertson frames these findings as associations, not proof that any page trait causes a citation. A prior Ahrefs test found cited pages were roughly three times as likely to contain JSON-LD, yet adding JSON-LD to already cited pages did not significantly change citations. The same caution applies here.
Start by auditing whether your product or service pages answer the buyer’s core question in the first block. Then run a controlled before-and-after test on a direct answer before rolling out sitewide changes.
Source: Search Engine Journal
