The existence gap: Chinese models mention brands 30.6 points more often on the same English questions
1,909 English-only questions, six models, 30 brands: Chinese models mentioned brands in 88.9% of cases versus 58.3% for international models — driven by training-data geography, not language.
Key numbers
- Study design: 1,909 pure-English questions across six models (GPT-4o, Claude, Gemini, Qwen3, DeepSeek, Doubao) and 30 brands.
- Chinese LLMs showed 30.6 percentage points higher brand mention rates than international LLMs (88.9% vs 58.3%, p<.001).
- The disparity persists on identical English questions, indicating that training-data geography — not the query language — drives the effect.
- Case study: a collaboration-platform brand scored a 65.6% mention rate in Chinese models and 0% in international models (p<.001): a brand absent from the training corpora effectively does not exist to AI.
Method and sample
arXiv paper “Cultural Encoding in Large Language Models: The Existence Gap in AI-Mediated Brand Discovery” (2025-12-30): 1,909 pure-English queries run across six models to compare mention rates for 30 brands, with a single-brand case study isolating the same-language, different-corpus effect.
What it means for brands
Whether AI knows you exist depends on whether your brand lives in the corpora it saw. For international markets, good English is not enough — your brand facts have to reach sources that the target market's models actually read, or you vanish the moment the engine changes.
How SourceSight handles it
SourceSight covers five international and five Chinese engines, so the same brand can be compared across both groups; multi-market projects are grouped together so a brand is not judged by one market alone.
Limits
The study covers 30 brands and 1,909 English prompts rather than a random sample of all brands, and it observes mention-rate differences without being able to attribute them to specific corpora inside the models.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
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.
Sources · 2026-10-06Incumbent 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.