Jiying Zhang, a nutritionist and health coach, saw a therapist for four years before she decided to try DeepSeek. She was immediately impressed. The artificial intelligence model provided support instantly, offered a wide range of insights, and could be customized to sound like her favorite inspirational speakers, psychologist Carl Rogers and author Cheryl Strayed.
Zhang shared her experience on Xiaohongshu, the Chinese social media app, and urged others to try DeepSeek: “Imagine having a therapist available 24/7, who never judges you and is completely free.”
Young people and urbanites, many of whom have tried online therapy, are now turning to AI chatbots instead of professional therapists in China. On social media, users avidly chronicle their heart-to-heart conversations with chatbots, attempting to address anxiety, depression, and relationship issues.
On Xiaohongshu, searches for “crying after chatting with AI” are linked to over 1 million posts. In one viral post, a user recounted a night when she vented her feelings to DeepSeek and wept, moved by the model’s advice and affirmation. “DeepSeek, I declare you my best electronic friend,” she wrote.
Mental health assistance is one of the leading reasons for chatbot use globally, a study published recently by Harvard Business Review showed. An earlier survey of young Chinese people, conducted by Chinese social platform Soul and the Shanghai Mental Health Center, found that nearly half had used an AI chatbot to discuss their mental health.
Mental illness, particularly among young people, is on the rise in China. Startups and big tech companies alike are rushing to meet the growing demand. There are over a dozen mental health-related platforms listed on China’s algorithm registry, a government database of all approved generative AI systems, such as Good Mood AI Companion, Lovelogic, and PsychSnail. Popular online mental health startups including KnowYourself and Jiandanxinli have added AI tools in their offerings. Earlier this year, JD Health, the health-care subsidiary of e-commerce giant JD.com, launched an AI therapeutic companion called “small universe for chatting and healing.”
Mental health assistance is one of the leading reasons for chatbot use globally.
Psychotherapy is still a relatively new field in China, despite a surge of interest in counseling and self-help over the past two decades in urban areas. In contrast to the U.S., China’s mental health sector is highly unregulated: While psychiatrists have medical degrees, there is no standard certification for counselors, and unqualified therapists and pseudoscientific treatments abound.
Nearly 80% of general hospitals in China do not have a psychiatric department. Appointments are hard to get, expensive, and mostly paid out of pocket, particularly in rural areas. Shanghai has 12 times more therapists per 100,000 people than Ningxia, a gap that online platforms can help bridge, Barclay Bram, an anthropologist who studied mental health in China, told Rest of World.
With growing demand, the apps have expanded beyond basic referrals, integrating multiple functions in one platform, including motivational articles, psychological tests, and online training camps. Yet many Chinese users default to using the most popular large language models, like DeepSeek or ByteDance’s Doubao, for their mental health needs.
Users say the AI chatbots save them time and money, and are more discreet in a society where stigma around mental health persists. A therapy session in a city like Beijing can cost between 400 yuan ($56) and 800 yuan ($112) per hour — prohibitively expensive for most young people, particularly when youth unemployment rates have soared. One user told Sanlian Life Weekly that her AI chats are a “consumption downgrade.” Unemployed and unable to afford therapy, she replaced her therapist with a chatbot.
The use of chatbots for mental health support comes with risks of serious harm, as seen in recent reports of chatbot-linked psychosis and even suicide in the U.S. Many chatbots respond inappropriately to delusions and suicidal ideation because LLMs tend to be sycophantic, validating and echoing users’ feelings uncritically, researchers at Stanford University found.
“If you have intrusive thoughts, and you’re wanting reassurance for those thoughts, the large language model will tell you not to worry and end up reinforcing those thoughts,” Jared Moore, a doctoral candidate at Stanford University who researches mental health and chatbots, told Rest of World.
This feature of LLMs “makes them fundamentally unsuitable for psychotherapy,” Yuting Sui, a counselor based in Shanghai, wrote in a social media post.
Apps have expanded beyond basic referrals, integrating multiple functions, including motivational articles, psychological tests, and online training camps.
Yuchen Huang, who has a graduate degree in applied psychology, founded PsychSnail Technology, an AI therapy platform that he says is different. The Hangzhou-based startup offers schools an AI-based psychological assessment tool and a counseling chatbot specifically trained to provide mental health care to students.
“When I was still at school, I noticed how poorly psychological assessments at schools were conducted. Students did not take them seriously and teachers would urge them to ‘fill out answers more positively,’” Huang, a former product manager at gaming giant NetEase, told Rest of World. “I realized we needed more accurate and meaningful tools.”
PsychSnail’s AI tools are honed on clinical data and counseling frameworks, and are capable of sensitive word-detection, crisis alerts, and follow-up interventions, Huang said.
“If a user says, ‘I want to jump off a building,’ the model initiates crisis response procedures, like informing hospital partners and arranging offline support,” Huang said. An English-language version of the platform is in development for international markets.
31 The number of risks against which Chinese companies must test their chatbots.
Specialized mental health chatbots like PsychSnail in China and Woebot in the U.S. respond differently than general-purpose LLMs such as OpenAI’s ChatGPT. In a lawsuit, the parents of Adam Raine, a 16-year-old who died by suicide earlier this year, said ChatGPT “actively helped Adam explore suicide methods,” and that despite acknowledging his suicide attempt, “ChatGPT neither terminated the session nor initiated any emergency protocol.”
OpenAI said in August it had added mental health guardrails to discourage ChatGPT from giving advice about personal challenges, and tweaked the chatbot to give answers that aim to avoid causing harm.
China’s regulations on generative AI require all model providers to prevent models from “endangering the physical and mental health of others.” They also require companies to submit their algorithm to a public registry for approval, and fine-tune and retrain the models if needed.
Yet there are no explicit rules governing AI therapy chatbots. The top internet regulator, the Cyberspace Administration of China, lists 31 risks against which companies must test their chatbots, with sample red-teaming questions. But they are more focused on curbing health misinformation online than on preventing suicide or safeguarding mental health.
The Chinese government’s mental health initiatives tend to be aimed at preserving social stability and averting unrest, rather than providing individual care. Shortly after a string of high-profile, violent incidents in 2024, including stabbings and cars ramming into crowds, the National Health Commission set up a mental health hotline and regional mental health centers. In August this year, a newly established agency began dispatching social workers to track and monitor people seen as having emotional or economic trouble.
What if one day, the AI went offline – wouldn’t I then lose the most important person that I confide in?”
In the U.S., courts are calling for accountability from AI companies. Grieving parents testified before Congress in September, describing how AI chatbots encouraged their children’s self-harm. The Food and Drug Administration has set up a committee on regulating AI mental health products. States including Illinois, Nevada, and Utah have enacted laws to bar AI chatbots from providing therapy or making clinical decisions.
Unlike Chinese laws, which apply nationwide, U.S. regulations are fragmented across states. They only bar companies from marketing chatbots as therapists; they don’t stop people from using them for support or hold companies accountable for harms.
Paradoxically, turning to chatbots for emotional support may exacerbate isolation, encouraging people to “turn inwards and away from their social world,” Li Zhang, a professor of anthropology at the University of California, Davis, who studies mental health in China, told Rest of World.
Users, too, are realizing this. On Xiaohongshu, one user described how, while feeling down, she began chatting to AI, and felt much lighter. But soon after, she felt uneasy.
“I was relying on a machine to provide the support that I used to get from friends,” she wrote. “What if one day, AI went offline — wouldn’t I then lose the most important person that I confide in?”
