The largest dataset on what earns an AI Mode citation puts domain traffic above every other factor, and large sites earn roughly three times the citations of small ones.

Generative UI builds a custom visual layout, interactive tool or simulation inside the answer rather than writing a paragraph. It launched in AI Mode in November 2025. Google's own wording on 19 August is that "Generative UI capabilities have launched globally in English in AI Mode and have started rolling out as well in AI Overviews." The announcement is framed around study tools, alongside practice quizzes, a Lens step-by-step mode, notebooks in AI Mode and custom file creation, so the first wave is educational queries rather than commercial ones. Search Engine Journal's read is that sites built around calculators, converters and comparison tools face the most direct overlap as it reaches AI Overviews, which is the larger of the two surfaces.

The method is the largest published on this question: 2,328,533 pages across 295,485 domains in 20 niches, citation frequency taken from over 500,000 prompts, modelled with XGBoost and read with SHAP so each factor gets a comparable weight. The results, in the vendor's framing, are that "domain-level authority and brand visibility dominates."

The numbers. Domain traffic carries a SHAP value of 0.63 for AI Mode and 0.62 for ChatGPT, the highest factor in both. Sites under 2,700 monthly visitors average about 2.4 citations; sites over 1.16 million average about 6.4, roughly a threefold gap. Referring domains split the two engines apart: 0.56 for AI Mode against 1.21 for ChatGPT, so backlinks matter about twice as much to ChatGPT. Content length runs the other way, 0.20 for AI Mode against 0.06 for ChatGPT. Sites with fewer than 300 referring domains average about 2.5 citations against about 6.8 for sites over 24,000, and pages under 500 words average about 4.1 against about 5.1 for pages over 2,300.

Two limits worth stating plainly. SE Ranking sells an AI visibility product, so this is a vendor study, and the split it reports happens to argue for buying more of the SEO work its tools measure. More importantly, it measures which pages get cited, not which businesses get named. Those are different events. A coach can be recommended by name in an answer that cites a directory page they do not own, and a large publisher can be cited constantly without ever being recommended to anybody. That distinction is the whole reason the free check measures naming.

Across roughly 300,000 domains, 10.13% had an llms.txt file, so nine in ten sites still do not. Citation frequency showed no significant correlation with having one, and when the researchers dropped llms.txt from the predictive model the model got more accurate, which is what a variable carrying noise rather than signal does. SE Ranking's own conclusion: it "doesn't seem to directly impact AI citation frequency. At least not yet."