Chinese AI and brokenhearted users

Alongside the rapid growth of AI globally, there has also been a rise in romantic AI relationships. A survey from late last year found that 28% of Americans have an intimate or romantic relationship with AI. This is also an issue in China, where the population continues to tumble, falling for its fourth straight year in 2025; the nation’s annual births have nearly halved from a decade ago in 2016, when China dropped its decades long one-child policy.

China’s fix? No more AI relationships.

Last week the country enacted new rules mandating tech companies disable features that draw users into emotionally dependent relationships with AI chatbots. The regulations ban the encouragement of emotional dependance with AI or exclusive virtual relationships, and specifically, AI companies cannot encourage users to withdraw from real-life relationships. These changes come after Tencent Research Institute’s April survey of 18–40-year-olds in China revealed over 70% of users have established an emotional connection with AI companions and more staggeringly, 56% of people turn to AI in situations that are “difficult to talk about”, compared to only 14.4% who prefer confiding in a human.

Its long been thought that China’s AI models have lagged six to eight months behind their American counterparts. Last week this changed, with the unveiling of Moonshot AI’s KIMI K3. The Beijing-based, Alibaba-backed startup revealed a model that outperforms Anthropic’s Claude Opus 4.8 and Open AI’s GPT-5.5, and the cherry on top - the model costs a fraction of its US equivalents. The announcement pushed anxieties over China’s place in the AI race to new heights, triggering a global tech sell-off, with chip stocks experiencing their worst week in 15 months.

China’s population has embraced AI in a very different way to western audiences. An AI sentiment survey from May of this year, found that 69% of western audiences say they trust AI-generated content less than human-generated content. Much of the AI-generated video we see in our feeds is usually funny, outrageous or clearly AI made, which delivers a quick laugh but doesn’t create deep, long-lasting engagement. China’s audience doesn’t experience AI in the same way. In China, AI-generated video content doesn’t detract from narrative, it enhances it. Chinese social media platforms such as Xiaohongshu (Rednote) and Douyin (TikTok) favour AI content, and it often replaces human-generated content entirely, instead feeding audiences specific AI videos that fit their algorithmic persona. Yet, China ranks 1st as the country with the highest media trust.

I visited a high-profile Shanghai-based AI media company, Ying Hai Drama Studio (which is almost impossible to search for on the western internet), that creates AI generated short and long-form videos for social media, often in the style of movies. The company claimed audiences on social media enjoy scrolling through AI generated movies that last up to two hours, split into 1–2-minute clips. This means the viewer scrolls through 45-120 short form video parts of a longer film which is storyboarded, scripted, designed, and generated entirely by AI. They base each film off a combination of current trends and the engagement levels of previous films. Interestingly, their current AI models can only generate 15 seconds of video at a given time, leaving AI engineers having to generate up to 480 clips (for a two-hour production), which then requires AI to stitch them all together. A time-consuming endeavor. Though, compared to the usual one to three year timeline to produce a film of such length, this is considerably more efficient and less labour demanding than human-made content, and that’s what much of this AI use boils down to - efficiency.

China is a vastly different (and larger) market for media consumption than Australia. But in a week where Nine Media has announced 30 journalists' jobs will go, perhaps there are some lessons for Australia - AI hasn’t replaced the need for authentic, human generated content for Australian audiences, but it can maximise efficiency in the way journalism is distributed. The ABC may be ahead of this curve: this month it announced the use of AI to turn regional radio news stories into online articles, as well as negotiating with Anthropic to incorporate their audio content into Claude AI models, according to ABC Managing Director Hugh Marks, pushing AI to play a supporting rather than a central role.

While AI girlfriends are breaking hearts and slop content isn’t going anywhere in a hurry, it does serve as a reminder that journalism holds special value as a form of media almost entirely built from human-connection. AI has the power to further advance how and when this connection reaches audiences.

References 

Edelman. (2026). 2026 Edelman Trust Barometer (p. 32). https://www.edelman.com/trust/2026/trust-barometer 

Lozo, D. (2026, July 16). China says dump your bot and go meet someone. Morning Brew. https://www.morningbrew.com/stories/china-says-dump-your-bot-and-go-meet-someone 

Salas, M. (2026, January 14). Artificial Romance: A Study of AI and Human Relationships. Vantage Point Counseling Services. https://web.archive.org/web/20260114162559/https://vantagepointdallascounseling.com/research/artificial-romance-a-study-of-ai-and-human-relationships/ 

Spero, M. (2026, May 15). 67% of People Consuming Online Content Are Spotting Misleading Info from AI. Pangram. https://www.pangram.com/blog/ai-sentiment-survey 

Tencent Research Institute. (2026, April 16). Chinese young netizens’ views and usage of AI-powered social media. 36Kr. 36Kr. https://36kr.com/p/3767898633519622