Analysis: China’s AI Companions Struggle to Turn Intimacy Into Industry

[para. 1] The global boom in artificial intelligence (AI) has influenced various facets of human life, with companionship emerging as a new frontier. In China, while there is growing interest in AI-powered digital friends and partners, developers are struggling to create sustainable business models out of these applications.

[para. 2] According to QuestMobile, two of China's leading AI social apps, ByteDance Ltd.'s BagelBell and MiniMax's Xingye, experienced a more than 12% decline in monthly active users year-on-year as of June. Sector-wide, Chinese AI social apps had 71.9 million monthly users in August 2024—a figure that pales in comparison to the over 650 million users of AI-powered search engines and assistants. Moreover, growth is stagnating, with a slight decrease from the previous month.

[para. 3] The competitive momentum in this sector is undermined by several issues: regulatory risk, technological constraints, and low user engagement. A government crackdown on AI-generated content, especially anything harmful to minors, raises risks for operators. Users also complain that current chatbots are too formulaic and emotionally distant, leading to low user retention, which threatens the subscription business models these apps rely on.

[para. 4][para. 5][para. 6][para. 7] Since 2024, the market for AI emotional companion apps in China has grown and can be divided into two categories: role-playing chatbots (e.g., BagelBell and Xingye) and apps featuring AI personas modeled after ideal romantic partners or celebrities. These apps mainly attract young women (ages 16–25), drawing inspiration from popular domestic romance games. For instance, the mobile game "Mr Love: Queen’s Choice" became a cultural phenomenon after its 2017 launch, garnering over 7 million App Store downloads in its first month. Over 70% of users were under 24, and it earned $56 million in iOS revenue in China in 2018.

[para. 8][para. 9] AI companion apps use varied character traits to build engagement. Trend-driven marketing on social media initially attracts users, but long-term engagement is hard to secure. New users need numerous interactions to be convinced of privacy and AI authenticity. The main challenge developers face is crafting AI personalities convincingly "human"—balancing flirtation and emotional ambiguity—something that current models struggle with due to their logical, less emotionally nuanced language.

[para. 10][para. 11] Further issues arise when bots forget past conversations or mix up details, which breaks the illusion of companionship. As a result, user retention is poor, undermining the core subscription model. Practical constraints include strict compliance demands and a largely young, low-spending user base.

[para. 12][para. 13][para. 14] The regulatory environment is strict. Virtual social apps are monitored because of their history with scams and inappropriate content. Regulatory focus has increased on age verification and content safety. For instance, in June, Shanghai authorities called in the operator of Zhumengdao for sexually suggestive content, demanding improved moderation.

[para. 15][para. 16][para. 17][para. 18][para. 19][para. 20][para. 21] Companion apps now moderate both user input and AI output. Preset, human-curated responses address sensitive matters; some apps also feature "psychological safety zones" with calming features if users express distress. Challenges similarly affect AI-powered toys, which must avoid inappropriate values and respond emotionally rather than mechanically to children.

[para. 22][para. 23][para. 24] National AI expert Guan Ming notes that, over time, AI’s tendency to please users could inadvertently worsen negative emotions. Notably, the suicide of a 14-year-old who interacted with an AI companion underscores the potential dangers. As AI continues in emotional roles, reported safety incidents rise, highlighting the need for better moderation and safeguarding systems.