An analysis of Baidu, Bilibili, Bing, Douyin, Weibo, and other Chinese search platforms finds 60K censorship rules, implemented differently by each service
This report has an accompanying FAQ. — Key findings — Across eight China-accessible search platforms analyzed — Baidu …
Context & Ripple Effects
This is the latest entry in a string of Citizen Lab investigations documenting how Western platforms behave inside China's information system: earlier reports found Bing surfacing Chinese state media content in its top US results on human rights and COVID origins, and Microsoft quietly hiding autofill suggestions for politically sensitive Chinese names in US Windows search before blaming a technical error. The new analysis scales up from single incidents to a full inventory — roughly 60,000 censorship rules across eight China-accessible search platforms, including Bing as the lone Western entrant.
The report also lands next to Beijing's own partial transparency move: the internet watchdog's published registry of algorithms running inside apps like WeChat, Taobao, and Douyin. Between the regulator's algorithm list and the Citizen Lab's rule inventory, there are now two independent maps of the same enforcement machinery — one official and coarse, one external and granular.
First-order effects
- Microsoft now appears in three separate Citizen Lab findings — the autofill episode, the state-media results report, and this rule inventory — making Bing the most-documented Western participant in China's search censorship apparatus.
- Chinese platforms including Baidu, Bilibili, Douyin, and Weibo learn that their filtering differs measurably from peers': the same query returns different results depending on which service is used, because each implements the rule set its own way.
Second-order effects
- With per-platform implementations documented side by side, compliance turns comparative — any one service's blocking behavior can be benchmarked against the other seven, raising the reputational cost of censoring visibly more or less than rivals.
- Pairing the CAC's algorithm registry with this rule inventory gives regulators an audit trail they could use to enforce consistency across platforms, while giving independent researchers the methodology to track enforcement drift over time.
Third-order effects
- China's draft AI rules extend content restrictions to chatbot interactions, so the rule sets measured here are the likely substrate for generative systems — making search-platform measurement the baseline technique for auditing what Chinese AI services will say.
- Fragmented implementation atop a shared rule set implies censorship operates through platform-level discretion rather than a single national filter, meaning accountability depends on external measurement like this instead of official disclosure.
The trend: Content control in China is evolving from opaque, per-platform moderation into a measurable, comparable censorship infrastructure that outside researchers can audit service by service.