Context
A large-scale empirical study of Chinese-language generative search analyzed eight web/app interfaces, 614 queries and 160,860 cleaned citation-level records, finding that entities present in citation pools were selectively surfaced and that source sets differed systematically between app and web interfaces.
What changed
The study provides quantitative evidence that citation-pool admission and answer-level entity exposure are distinct stages, and that interface type can materially change source selection within the same platform.
Why it matters
AI visibility measurement should not collapse retrieval, citation and entity exposure into one metric. Interface should also be treated as an experimental variable when auditing generative-search visibility.
Visibility OS interpretation
Visibility OS interprets the findings as evidence for a staged visibility model: source retrieval/citation creates an opportunity for exposure, but does not guarantee that an entity is surfaced in the generated answer. This interpretation is limited to the study scope.
What remains uncertain
How stable are these effects over time and across languages? Do comparable web/app differences occur in ChatGPT, Google AI Mode, Perplexity, Claude, Gemini or Bing/Copilot? Which source and entity features causally affect transition from citation-pool presence to answer exposure?