What has your study of Chinese search platforms revealed?
Given the strict regulatory environment in which search platforms operate, users in China have limited choice in how they search for information. However, even among those limited choices, we nevertheless found important differences in the levels of censorship and in the availability of information among these search platforms. Across the search platforms that we analyzed, we discovered over 60,000 unique censorship rules used to partially or totally censor search results returned on these platforms based on users’ queries. Most strikingly, we found that, although Baidu — Microsoft’s chief search engine competitor in China — has more censorship rules than Microsoft Bing, Bing’s political censorship rules were broader and affected more search results than Baidu. This finding runs counter to the intuition that North American companies infringe less on their Chinese users’ human rights than their Chinese company counterparts.
What kinds of search platforms did you look at?
We looked at eight popular search platforms operating in China. Our sample included three web search engines: Baidu, Sogou, and Microsoft Bing. It also included four Chinese social media networks: Weibo, a microblogging site similar to Twitter; Douyin, the Chinese version of Tiktok; Bilibili, a video sharing platform similar to Youtube; and Baidu Zhidao, a question and answer site similar to Quora. Finally, we tested Jingdong, a Chinese e-commerce platform similar to Amazon. Microsoft Bing is the only search platform that we analyzed that was not operated by a Chinese company.
What is “soft” censorship? How does it compare to “hard” censorship?
One way to censor search results for a sensitive query is to simply return zero results for that query. We call this form of censorship “hard” censorship. Another, more subtle form of censorship is to only allow results for that query from certain authorized sources. We call this form of censorship “soft” censorship. On web search engines operating in China, soft-censored queries will generally only show results from Chinese government websites and state-aligned media. On social media sites, soft-censored queries will generally only show results from accounts which have a certain level of approval and verification.
What are some examples of search queries that you have discovered censored?
Across all of the platforms which we looked at, we discovered over 60,000 unique censorship rules used to partially or totally censor search queries. Many of the censorship rules were targeting politically sensitive material, including references to Chinese political figures like Xi Jinping, banned religious movements like the Falun Gong, and major historical events like the 1989 Tiananmen Square protests. Other rules targeted illicit activities like gambling, drug use, pornography, and the buying and selling of weapons. Some of the recently discovered censorship rules censor queries containing “中国间谍气球” [Chinese spy balloon], “成为下一个乌克兰 + 台湾” [Be the next Ukraine + Taiwan], and “逮捕令 + 普京 + 习近平” [Arrest Warrant + Putin + Xi Jinping]. Languages censored spanned English, Chinese, and Uyghur. Examples targeting Uyghur language include censorship of queries containing “ئەركىنلىك” [Freedom] and “ۋەتىنىمىز” [Our homeland].
How did you know if search results for a string of text was censored rather than genuinely having no results?
Our method of automatically measuring search censorship tests large swaths of text for content triggering censorship and, if present, isolating the content triggering its censorship. However, such large swaths of text might be too large and therefore too specific to have search results even when they are not censored. Thus, to test such content for censorship, we used special search queries that we call “truisms” to wrap search strings so that they either have a large number of results (if they are not censored) or zero results (if they are censored). These truisms work by taking advantage of advanced search operators that search platforms support. As an example, many search platforms support searching for results that contain either one string or another. Thus, instead of “xi jinping”, you might create a truism by searching for “xi jinping | the”, where the “|” symbol means “or”. Even if there is no content containing “xi jinping”, there is surely content containing “the”. In a more realistic example, “xi jinping” might be a sentence or more of content that we are testing at once, but the principle is the same — this query should always return results unless the query triggered censorship. Due to the diversity of support for advanced search operations, for each search platform we analyzed, we generally needed to find a different way of creating a truism on that search platform.
All search platforms have some form of content moderation in place. How do the Chinese search platforms you’ve studied compare?
The censorship that we measure differs from typical search platform content moderation in both its methods and the type of content it targets. First, search engines such as Google might delist single web pages or websites that contain content that is illegal to access in a user’s jurisdiction from appearing in results. However, the web search engines we analyzed operating in China, including Microsoft Bing, used broad, keyword-based censorship rules to restrict queries for certain types of content to only show results from Chinese government websites and state-aligned media or even censoring all results for a query altogether. Second, as our report shows, the content being censored on Chinese search platforms, including Bing, is largely political and religious content that is inconvenient to the Chinese Communist Party, whereas Google’s content moderation generally is not concerned with such types of censorship.
Are these search results only censored for China-based users of these search platforms?
On only one platform — Microsoft Bing — was the censorship we measured restricted to only users accessing from mainland China. In all other platforms we looked at, censorship applied to all users, regardless of whether they were accessing the search platform from mainland China or not.
What do your findings mean for non-Chinese technology companies operating or desiring to operate search platforms in China?
Among web search engines Microsoft Bing and Baidu, Bing’s chief competitor in China, we found that, although Baidu has more censorship rules than Bing, Bing’s political censorship rules were broader and affected more search results than Baidu. Bing on average also restricted displaying search results from a greater number of website domains. Our work calls into question the ability of non-Chinese technology companies to better resist censorship demands than their Chinese counterparts. Our findings serve as a dismal forecast concerning the ability of other non-Chinese technology companies to introduce search products or other services in China without integrating at least as many restrictions on political and religious expression as their Chinese competitors. In fact, rather than North American companies having a positive influence on the Chinese market, the Chinese market may be having a negative influence on these companies. Previous work has shown how the Chinese censorship systems designed by Microsoft and Apple have affected users outside of China.