Centralised Diversification: China’s State Media Framing of AI Governance for International Audiences on YouTube and TikTok

Democracies confront cross-border political communication on social media platforms, including messaging from authoritarian states. This paper analyses how Chinese state media communicate internationally about artificial intelligence (AI) and AI governance on YouTube and TikTok. AI functions as both a policy domain and a symbolic resource: external messaging advances preferred governance principles, manages reputational risks linked to surveillance and information control, and positions China as a source of technological capacity and regulatory expertise. By normalising specific problem definitions and governance vocabularies, this communication can shape how publics interpret disputes over data and platform regulation. I develop the concept of centralised diversification. Central authorities stabilise the core claims of AI governance discourse, consistent with the pursuit of huayu quan (discourse power), while state media vary frames and presentation to fit platform affordances and audience expectations. Influence operates through agenda setting and framing, not only misinformation, linking strategic narrative scholarship to international norm competition in technology governance. Two questions guide the study: how do frames differ between content targeting Global South and Global North audiences, and despite calibration, how closely do outward narratives track China’s domestic AI policy discourse? I expect a disciplined policy line emphasising state capacity, orderly development, and governance legitimacy, alongside predictable secondary variation. Development, infrastructure, and "AI for social good" cues should be more salient in Global South targeting, whereas safety, ethics, standards, and competitiveness cues should be more salient in Global North targeting. I also expect TikTok posts to rely more on short-form human-interest storytelling and visual symbolism, while YouTube content devotes more space to institutional authority and policy explanation. Empirically, I compile a multilingual corpus of outward-facing AI-related posts from China Global Television Network (CGTN, multiple languages), Xinhua, China Daily, People’s Daily, Global Times, CCTV+ (CCTV Video News Agency), and China News Service on YouTube and TikTok (January to September 2025; English, Spanish, French, Chinese). Audience labels draw on pre-period channel targeting metadata and are reassigned conservatively at the post level when distribution cues indicate cross-audience reach, enabling comparisons by audience segment, platform, and quarter. I track language choice and cross-posting to assess coordination. Frames are estimated using a contextual topic model and mapped onto a short frame codebook specified in advance. To connect messaging to state strategy, I introduce a Policy Echo Score measuring semantic similarity between each post and a reference corpus of Chinese AI policy texts (2017 New Generation AI Development Plan, subsequent ethics guidance, 14th Five-Year Plan). I estimate frame prevalence by audience, platform, and quarter and test whether Policy Echo predicts tighter frame discipline. I validate results through bilingual spot checks, split-sample stability tests, and robustness checks to translation choices and vocabulary dilution. The paper clarifies how China calibrates AI governance narratives across audiences while remaining anchored in central policy language, with implications for democratic resilience and platform governance debates. It also offers a reproducible approach to linking multilingual platform framing to domestic policy discourse.