What Chinese generative search actually cites: 8.3% brand selection, and app vs web cite different sources
Across 160,860 citation-level records, the same platform surfaced systematically different source sets on its app and web interfaces.
Key numbers
- Scale: four mainstream Chinese platforms across web and app — eight interfaces, 614 queries, three replications per query-platform-interface combination; 214,119 raw records cleaned into 160,860 citation-level records.
- Brands appear in the citation pool but are selectively surfaced: the overall brand-selection rate was 8.3%, and only 12.4% of retrieved sources containing contact information contributed that contact information to answers.
- In predictive models, content fit, cross-source occurrence count and semantic role mattered most, while a composite Baidu quality score was not the leading predictor for any outcome examined.
- Recency: among cited pages with publication dates, fitted half-lives were about 39 days for high-timeliness queries and 68 days for low-timeliness queries.
- About 13% of brand exposures could not be matched to the contemporaneous citation pool, about 71% of contact-information exposures could not be matched to crawled body text, and app versus web source sets differed systematically.
Method and sample
arXiv paper “What Do Chinese-Language Generative Search Engines Cite and Surface? A Large-Scale Empirical Study” (2026-07-17): a controlled design over eight platform interfaces, 614 queries and three replications per combination, building 160,860 citation-level records from 214,119 raw records to analyse citation behaviour, source attribution, entity exposure and cross-interface consistency.
What it means for brands
In Chinese-language search, being retrieved and being written into the answer are different events: even inside the citation pool only 8.3% of brands get surfaced, and contact details usually do not make it into the answer. And app and web are not one entry point — their source sets differ, so checking one alone gives the wrong answer.
How SourceSight handles it
SourceSight runs one project per market and reports mentions, position and cited sources per engine; Chinese projects ask in Chinese and get a Chinese-market off-site list rather than an extrapolation from overseas data.
Limits
The study covers four Chinese platform interfaces and query distributions that may differ from your industry; the half-lives are fitted values, not an expiry rule for all content.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
Incumbent advantage: with identical specs, a well-known brand is recommended 100% of the time
Three experiments on skincare: identical specifications produce near-total dominance for known brands — and less than a +0.1-star edge for a competitor breaks it.
Sources · 2026-10-06Does AI judge content or clout? A causal test of authority bias
Hold the title and abstract constant, change only author prestige, venue and citation counts, and eight models change their recommendation — while claiming they do not.