The Chinese AI source map: Doubao leans on Toutiao and Douyin — and every engine cites Zhihu
Posting the same content everywhere averages a 10.6% citation rate; rewriting it per platform averages 38.3% — a 3.6× gap.
Key numbers
- Zhihu is the only platform cited frequently by every Chinese engine: Kimi ~18%, DeepSeek ~15%, Doubao ~12%, Qwen ~10% — all four are engines we monitor.
- Doubao leans on the ByteDance ecosystem: Toutiao ~35%; Douyin content reaches a 97.7% citation rate for travel-guide style consumer decisions; the same content is cited from Douyin in ~30% of app-side answers but only 11.85% on web.
- One piece cross-posted to 8 platforms averages a 10.6% citation rate; adapted per platform it averages 38.3%. Brands whose claims conflict across platforms see an 82% drop in the chance of being the first recommendation.
Method and sample
Compiled from a public industry verification report dated 2026-09-19 (data as of August 2026): it aggregates source tracking across 10 industries and 6 Chinese AI platforms, a 16,800-query test, and 100M+ citation records from several research teams. Only the four engines we monitor (DeepSeek, Kimi, Doubao, Qwen) are quoted below; measurement methods differ, so percentages vary between sources.
What it means for brands
There is no single Chinese checklist: Doubao starts with Toutiao and Douyin. DeepSeek, Kimi and Qwen have no captive content pool and reward overall content quality — but Zhihu is the shared lever across all four. Inconsistent brand claims across platforms get significantly down-weighted.
How SourceSight handles it
SourceSight builds the off-site list from your target market and business type — Chinese projects never get Yelp, overseas projects never get Zhihu — and orders it by who AI already cites, so the biggest gaps come first.
Limits
A secondary verification report, not peer-reviewed research. It also covers Yuanbao and Ernie, which we do not monitor — their numbers are not quoted here. Major platforms reworked source rules after March 2026, and percentages shift by time and by surface (app vs web), so treat these as direction rather than targets.
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.