Tencent has entered consumer search with a dedicated product rather than a chatbot feature. Yuanbao Search went live on November 11 at so.qq.com, and users of QQ Browser can also switch it in as their default engine. The company says it is web-only for now, with no client app planned. The product is built on the Yuanbao assistant's large language models and rebuilds the entire search pipeline — understanding, retrieval, ranking and answer organization — around generative AI rather than bolting an AI summary onto a classic results page.
The index claims are the headline. Tencent says Yuanbao Search runs on a hundred-billion-scale index covering the open web, WeChat official accounts, video account content, Tencent ecosystem material and academic literature. That WeChat reach matters: public-account articles have long been a walled garden that competing engines index poorly, and Tencent now owns both the content and the search layer on top of it.
Result presentation follows the pattern regulators and users have been pushing for. The engine prioritizes authoritative institutions and flags official websites, and it keeps advertising visually separated from answers, with a one-click option to collapse the entire ad block. Tencent says the product draws on 157 authoritative sources.
The academic search mode is the most distinctive feature set. Yuanbao Search summarizes paper conclusions in plain language while keeping the original sources indexed, supports one-click translation of English literature, and can read titles and abstracts aloud. Search input is multimodal — typing, speaking or photographing a query all work — and the results page carries an AI summary with an "AI mode" that allows consecutive follow-up questions by voice, text or image while preserving context.
Tencent also claims the product improves itself through a "recursive self-improvement" (RSI) mechanism that mines real usage and feedback to keep refining the search experience. That framing is notable in itself: the same RSI terminology that frontier labs use for model self-improvement is now showing up in consumer product marketing, applied to ranking and answer quality rather than raw capability.
The launch puts Tencent directly against Baidu's AI-transformed search, Alibaba's Quark, ByteDance's assistant-driven discovery and the global AI answer engines. Tencent's structural advantage is the WeChat content moat plus distribution across QQ Browser; the open question is whether a purpose-built product can convert Yuanbao's chat users into habitual searchers. All index and source figures come from Tencent's own announcement and have not been independently audited.
Comments (0)
Log in to join the discussion
Log InNo comments yet