What actually predicts brand mentions in Chinese AI search
Across 240 brands probed on DeepSeek, Qwen, Doubao, Kimi and Baidu ERNIE, encyclopedic presence was the strongest single predictor of brand mention rate: brands with a Baidu Baike entry were mentioned 3.1x more often than brands without one. On-site schema markup density showed no positive correlation and a mild negative one, which inverts the standard Western AEO checklist.
- Method
- 240 brands × 60 prompts × 5 Chinese engines, Simplified Chinese prompts, CN egress
- Sample
- 72,000 answers
Why Chinese engines need their own research
Almost all published AEO guidance is derived from Western engines trained on the open web. Chinese engines are trained on a different corpus with different trust hierarchies. Applying a Western checklist there is not conservative — it is wrong in specific, measurable ways.
Method
We selected 240 brands with meaningful Chinese-market presence across consumer, B2B and industrial categories. Each was probed with 60 prompts in Simplified Chinese across DeepSeek, Qwen, Doubao, Kimi and Baidu ERNIE, ten samples per prompt, collected from CN egress with genuine accounts on the gated engines.
For each brand we recorded: Baidu Baike entry (yes/no), Chinese Wikipedia entry, Zhihu answer volume, WeChat official account activity, Douyin follower band, schema.org density on the brand's Chinese site, and Chinese-language page count.
Predictors, ranked
| Signal | Correlation with mention rate | Notes |
|---|---|---|
| Baidu Baike entry | +0.61 | Strongest single predictor across all five engines |
| Chinese Wikipedia entry | +0.44 | Strongest on DeepSeek and Kimi specifically |
| Zhihu answer volume | +0.38 | Concentrated in B2B and technical categories |
| Douyin presence | +0.35 | Almost entirely a Doubao effect |
| Chinese-language page count | +0.19 | Weak, and flattens above ~200 pages |
| schema.org density | −0.08 | No positive effect; mildly inverted |
Baike is close to a gate. Brands without a Baidu Baike entry were mentioned in 9% of relevant prompts on Baidu ERNIE. Brands with one were mentioned in 31%. No amount of on-site content closed that gap in our sample.
Doubao is a different web. Brands with no short-video presence averaged a 4% mention rate on Doubao regardless of how strong everything else was. If your client sells in China and ignores Douyin, Doubao visibility is structurally near zero.
Kimi rewards depth over breadth. Brands with at least one long-form Chinese technical document outperformed brands with many short pages by 2.4x, consistent with Kimi's long-context ingestion.
The uncomfortable finding
Caveats
Correlation, not causation; brands with Baike entries are also larger and better known. Coefficients are computed across n=240 brands; per-coefficient significance and confidence intervals are available on request, and the −0.08 on schema density is not distinguishable from zero — we read it as "no positive effect," not as evidence of harm. We are running a follow-up on newly created Baike entries to test the direction of the effect and will publish it regardless of outcome.
Evidence
Raw answers, per-cell sample sizes, collection dates and parser versions behind every figure here are available to customers on request. We publish the method so the numbers can be checked rather than taken on faith.
See this for your brand
Every figure here comes from the same measurement behind our dashboard and API. Tell us what you want measured and we'll show you where your brand stands.