The question’s language is the switch — not the engine
Every engine recommends far more Chinese products when asked in Chinese. Ask the same question in English and Chinese brands all but vanish — even on China’s own engines. Which company built the AI matters much less than which language you type in.
| Chinese-brand share of recommendations | Asked in English | Asked in Chinese |
|---|---|---|
| Western engines (GPT · Claude · Perplexity · Grok) | 1.3% | 26% |
| Chinese engines (豆包 · 通义 · DeepSeek · Kimi) | 1.3% | 31.1% |
Share of named brands that are Chinese products. Both engine groups sit near 1.3% in English.
What this means for a brand selling abroad: if your buyers ask AI in English, you are invisible on every engine — including the Chinese ones. Showing up is a per-language problem, not a per-engine one.
Western SaaS dominates — even inside Chinese engines
Asked in Chinese, on Chinese engines, the most-recommended B2B tools were still Western. Chinese products rarely cracked the top of the list.
| Most recommended — Chinese engines | Answers naming it |
|---|---|
| HubSpot | 31 |
| Microsoft Teams | 27 |
| Notion | 22 |
| Slack | 16 |
| Figma | 16 |
| Asana | 15 |
| Jira | 15 |
| Sketch | 15 |
Count = number of answers (of 356 total) in which the engine named the brand.
The Chinese engines were the reliable ones
Counter to the usual assumption, the four Chinese engines returned a valid answer to every single query. The failures were all on the Western side — Grok’s search API worst of all.
| Engine | Origin | Failed to answer |
|---|---|---|
| Perplexity | West | 0% |
| DeepSeek | China | 0% |
| Kimi | China | 0% |
| 通义 Qwen | China | 0% |
| 豆包 Doubao | China | 0% |
| Claude | West | 12.5% |
| ChatGPT | West | 12.5% |
| Grok | West | 33.3% |
Share of queries that returned an error or timed out, across 384 total calls.
Chinese engines don’t show their sources
Western engines cite the web — they returned links to G2, Reddit, CSDN, Zhihu and vendor blogs. The four Chinese engines returned zero machine-readable citations. They give an answer; they don’t show where it came from. For anyone trying to earn visibility, that means the Western playbook — get cited on the sites AI reads — doesn’t transfer to Chinese engines unchanged.
| Most-cited domains — Western engines | Times cited |
|---|---|
| cnblogs.com | 27 |
| blog.csdn.net | 27 |
| learn.g2.com | 21 |
| zoho.com.cn | 18 |
| youtube.com | 17 |
| zhuanlan.zhihu.com | 16 |
| cloud.tencent.com | 14 |
| reddit.com | 14 |
| pcmag.com | 14 |
| hubspot.com | 12 |
Chinese engines (豆包 · 通义 · DeepSeek · Kimi) returned no citation URLs at all.
Method
Between the western engines (ChatGPT, Claude, Perplexity, Grok) and the Chinese engines (豆包 Doubao, 通义 Qwen, DeepSeek, Kimi), we asked 12 B2B buying questions — the best CRM, project tool, e-signature, help desk and so on — each in both English and Chinese, with web search enabled, two repetitions per cell. Brand mentions were detected by deterministic keyword matching against a fixed roster that mixes Western and Chinese incumbents in every category, so the numbers are reproducible rather than model-judged.356 of 384 cells returned a valid answer. AI answers are non-deterministic; treat these as directional, not precise. Measured July 2026 by Avowd.
Is your brand in the answer — in both languages?
Avowd measures how nine AI engines describe and recommend a brand across English and Chinese, then does the off-site work to change it. Run the free check, or ask for the full audit.