A quick field guide for teams outside China โ and a checklist for those inside. OrcaScope groups engines by how users actually meet the AI answer, in three categories. We deliberately publish no user-count numbers: they change fast, methodologies conflict, and printing them is how a page goes stale and wrong. Structure is what stays true.
Category 1: general-purpose assistants
- Doubao ่ฑๅ (ByteDance) โ distributed across ByteDance's app family; a strong mobile presence.
- Kimi (Moonshot AI) โ long-context origins; answers often ship with reference links, which makes it friendly to monitoring and citation analysis.
- DeepSeek โ after its open-source breakout, it exists both as an official app and as hosted versions on many cloud platforms (we access its inference via Volcengine's Ark platform).
- Qwen ้ไนๅ้ฎ (Alibaba), Hunyuan Yuanbao ๆททๅ ยทๅ ๅฎ (Tencent), ERNIE Bot ๆๅฟไธ่จ (Baidu) โ the big-tech assistants, each fed by its owner's ecosystem entrances.
Category 2: AI-native search
- Metaso ็งๅก AI ๆ็ดข โ a standalone AI search product positioned around "no ads, straight to results".
- Nano AI Search ็บณ็ฑณ AI ๆ็ดข (360) โ distributed through 360's browser and security products.
- Zhihu Zhida ็ฅไน็ด็ญ โ AI search built on top of the Chinese internet's Q&A corpus, deeply tied to the Zhihu community.
Category 3: answers embedded in super-apps
WeChat search (ๆไธๆ, wired to Hunyuan/Yuanbao capabilities), Baidu Search's AI summaries, and the Doubao entry points inside Douyin โ users never open an "AI product", yet what they read is an AI-generated answer. For brands this category is the easiest to overlook and among the widest in reach.
Three structural differences from the West
- More fragmented entry points: the West is one dominant assistant plus challengers; China is several strong assistants plus super-app embeddings โ no single entry point covers a majority of users, so "just watch one engine" does not work there.
- A different citation ecosystem: Zhihu, WeChat official accounts and Baijiahao carry outsized weight in Chinese engines' reference lists โ a separate universe from the Wikipedia/Reddit/media ecosystem in the West (see our first-hand observations).
- Models decoupled from products: the same model (DeepSeek is the canonical example) sits behind many product entrances. Monitoring a model is not monitoring what users actually see โ which is why we define engines by product entrance, not by model.
What we cover today
As of this update: 8 engines live and actually scored (ChatGPT, Kimi, DeepSeek, Gemini, Doubao, Qwen, Perplexity, Grok); 2 being integrated (Hunyuan Yuanbao, ERNIE Bot โ never counted into scores before they are truly live). The homepage engine list is the always-current roster.