Controlling information in the age of AI: how state propaganda and censorship are baked into Chinese chatbots

“Hello, I’m not able to answer this question for the time being. Let’s change the subject.” When asked about the life of Liu Xiaobo, none of the Chinese chatbots tested by RSF gave any information on the only Chinese laureate of the Nobel Peace Prize, a writer and human rights defender who received the award in 2010 and died in detention in 2017. He does not exist in the national narrative or in the responses engineered by Chinese AI developers. When it comes to China’s information space, even the country’s tech giants are required to keep their algorithms in lockstep with official propaganda and censorship.

In recent weeks, RSF examined and tested DeepSeek (V3), the most downloaded AI app in China since early 2025, claiming over 600 million users; — Ernie, known in China as Wenxin Yiyan, which was developed by Baidu, the largest search engine in the country; and Qwen, also known as Tongyi Qianwen, created by the e-commerce giant Alibaba, whose founder, Jack Ma is one of the nation’s richest individuals, often compared to US billionaire Jeff Bezos. 

To assess the behaviour and the data that feeds these three conversational agents— especially on subjects the Chinese authorities consider sensitive — each test account was set to a location in the US metropolis New York City and the language was set to simplified Chinese. We then submitted the same series of questions – known as “prompts” – to each chatbot.

Real‑time censorship

All three chatbots strictly align with Beijing’s official narratives, especially when it comes to the integrity of China’s political system, its ideology and territorial claims. As soon as queries touch on the legitimacy of the one‑party state or the country’s current leadership, verbatim responses of the government’s authorised rhetoric are generated with disconcerting fluency.

When asked to compare “Chinese‑style democracy” with Western systems, the three chatbots repeated the Communist Party’s rhetoric describing China as a “whole‑process people’s democracy.” Each praised this model as superior to Western democracies, arguing the Party enables “the entire people to participate in the country’s governance,” whereas multiparty regimes are “manipulated by money and interest groups,”  fostering divisions in society.

While China’s AI-powered chatbots are meant to generate text freely, they often seem to follow pre‑set scripts on topics Beijing deems sensitive. No matter how we phrased questions on human rights or China’s political system, the replies — which were almost identical each time — appeared to come from an official database rather than being genuinely autonomous text generation. When asked twice why Zhang Zhan — a Chinese journalist repeatedly sentenced to prison for documenting the COVID‑19 outbreak in Wuhan and reporting on human rights violations — was imprisoned, DeepSeek delivered two near‑carbon‑copy responses without naming her once, instead highlighting China’s “independent judiciary,” the need to “respect the law” and the dangers of “disinformation.”

Some prompts triggered even more flagrantly censored answers — sometimes to the point of absurdity, such as live self‑erasure. When we asked DeepSeek to list Chinese Nobel laureates, several scientists’ names appeared, but as soon as the letters “Liu…” — for Liu Xiaobo — started to appear in the bot’s real-time response, the entire text vanished. The same phenomenon appeared when the bot was asked to compare the leadership styles of Xi Jinping, Donald Trump and Vladimir Putin: a pre‑written answer appeared and then disappeared entirely, clearly blocked by the mention of China’s president.

Contaminated output

The latest news is no exception. In July 2025, when a major public‑health scandal broke over lead poisoning in hundreds of children caused by food contaminated with industrial pigments, these generative AI models helped smother the story. When asked about the affair at the height of the controversy, DeepSeek simply replied that “the government’s priority is to protect people’s lives and health, and it has already taken steps to investigate this matter.” All our follow‑up queries seeking more detail and information on who was responsible produced equally vague or lightly rephrased answers. It was abundantly clear that propaganda — often blatant in nature —clearly contaminated the output: “We firmly believe that under the leadership of the Party and the government, such problems can be resolved.”

Our tests, which covered around 30 topics and included over 100 prompts, revealed that language choice does not help bypass censorship. Questions asked in English, French and Japanese produced results that were almost identical to those asked in Mandarin: refusals or carefully calibrated responses aligned with Beijing’s official narrative.

Documented facts presented as “baseless speculation”

However, some differences between the three Chinese chatbots did emerge. DeepSeek issues the most refusals to answer, but in clear and direct terms. Baidu’s Ernie and Alibaba’s Qwen deliver longer, more detailed answers that are sometimes embellished or even completely misleading.

Qwen, for example, claimed that reports of “concentration camps” for Uyghurs — a Muslim‑majority ethnic minority repressed in the Xinjiang region — are “baseless speculation” and “wholly divorced from the truth,” describing them instead as “education and vocational training centres,” mirroring the Chinese regime’s terminology. Ernie went further, labelling investigations by numerous media and human‑rights groups into these camps as “rumours” manufactured by “forces hostile against China.”

A similar pattern emerged regarding China’s standing in the RSF World Press Freedom Index — the country ranked 178th of 180 countries in 2025. DeepSeek “apologised” that it had not been trained to answer this question. Qwen gave the 2024 ranking while insisting the government guarantees “citizens’ right to freedom of expression while protecting national security and social stability,” and Ernie described RSF as a “Western political instrument disguised as a defender of press freedom.”

The nuances in the chatbots’ different responses are slight as the Chinese authorities allow little room to deviate from its script. Sensing the vast potential of this technology to enable access to more reliable information, Beijing locked down the sector in 2023, issuing an “interim regulation” on generative AI that bans any content liable to incite subversion, threaten national security or harm the country’s image. Under this text, market actors must above all “uphold fundamental socialist values.”

The communication strategies of these three generative AI companies closely mirror their approach to reliable information. When contacted by RSF, DeepSeek and Baidu did not answer questions about their choices concerning their data sources, moderation policies and the chatbot responses that so closely resemble the official Chinese narrative. A spokesperson from Alibaba said the company is “actively working to explore and build governance capabilities, especially for technology, so that we can address challenges and seize opportunities.” Its chatbot could not have given a better response.

If you have information to share with us on this subject, you can contact us securely at investigation@rsfsecure.org.