Now it’s China’s experts who are gig workers training AI data

As a Shenzhen-based architect in her 40s, Cuicui has vast experience designing public stadiums, schools, and hospitals at an architecture firm. But over the past five years, declining government infrastructure spending caused a sharp drop in business. Her income halved. To keep paying her mortgage and cover her child’s extracurricular fees, Cuicui became a specialized AI trainer. At the end of her regular workday, she would log on to a data annotation platform called TalentsAI, and give an AI model step-by-step instructions on how to write up a building proposal or run profitability analyses. “It is just like teaching a child,” Cuicui told Rest of World. “Every task has to come from my actual work.” You have to keep evolving yourself to avoid getting completely replaced.”A Shanghai-based software engineer with two decades of experience who began taking AI training gigs in July. In the U.S., data companies like Mercor, Surge AI, and Handshake have already recruited tens of thousands of Hollywood filmmakers, lawyers, and scientists to train AI models on various white-collar tasks. These jobs are now emerging in China, as the government pushes for AI adoption across the economy. In June, China’s National Data Administration threw its support behind building “high quality data sets,” and called for more industry experts to be brought deeper into data annotation to raise the “knowledge density” of AI training data. It encouraged college graduates to take up annotation work as a source of flexible employment. Rest of World spoke with a cohort of expert-level AI trainers who, like Cuicui, requested anonymity to avoid being identified by employers, colleagues, or the training platforms they feed. They said the work is precarious, demanding, and lower-paid than in America. But as they face a tough job market and disruption from AI, the Chinese professionals say they have taken the gigs to make money as well as to participate in what they believe to be an industry of the future. “What can middle-aged people do?” AlibabaiAlibabaAlibaba, founded in 1999 by Chinese entrepreneur Jack Ma, is one of the most prominent global e-commerce companies that operates platforms like AliExpress, Taobao, and Tmall.READ MORE’s AI training platform, Siriser, which launched in January, is recruiting teachers, mechanical engineers, and music composers as AI trainers. ByteDance’s Xpert, launched in 2025, says it has recruited more than 50,000 experts, from writers to therapists. Other independent platforms, like TalentsAI and MeetChances, are hiring journalists, tax accountants, and headhunters as gig AI trainers. “In the AI era, what can middle-aged people do after getting laid off?” MeetChances posted on social network RedNote — Xiaohongshu in Chinese — in June. “Why not try becoming AI trainers.” In recent months, Chinese AI labs have narrowed their gap with Silicon Valley in building models that excel at general reasoning and coding. They are now ramping up specialized data collection, expecting AI work tools to be the next hot spot, Tiezhen Wang, a Sydney-based AI researcher and former head of the Asia-Pacific ecosystem at Hugging Face, told Rest of World. Tech giants have launched new AI office tools, such as Tencent’s WorkBuddy and ByteDance’s Doubao Work, which can generate PowerPoint slides and build websites. But getting AI to work in medicine, law, and finance requires more data. “Making models perform better at work will be a very strong differentiator in the future AI competition,” Wang said. Similar to their Western counterparts, the freelance AI trainers said they are told to come up with realistic work tasks, upload documents from their actual jobs, and write down their own thought process so AI can learn from it. Trainers are not allowed to use AI themselves. It usually takes a few hours to complete a task, which earns them 100 to 500 yuan ($15 to $74), they say. If I don’t pass my knowledge to AI, others will do that.”A Shenzhen-based AI trainer with a master’s degree in law who decided not to join a law firm. Trainers say they must design assignments that the models cannot yet solve on their own — that has become increasingly difficult as AI continues to improve. Gig trainers have also complained that their outputs are sometimes rejected by quality control teams, which means they don’t get any money for hours of work. Platforms including Xpert have stated in their user terms that trainers will not be paid if their submissions do not meet requirements or pass reviews. Siriser said trainers could make revisions or appeal if their submissions were rejected. Despite the challenges, data workers say training AI is a way to diversify their income in an uncertain economy. The youth unemployment rate, at 17.9% in July, is approaching the highest in three years, as a record number of college graduates entered the job market this past summer. A Shenzhen-based AI trainer, who requested to be identified by her last name, Liu, said she had a master’s degree in law but decided not to join a law firm. She wanted to avoid the long hours, low pay for junior employees, and a culture of sexual harassment she had been warned about. Liu works at a legal AI company during the day. After work, she takes AI training gigs at Moonshot AI, Tencent, and TalentsAI. She asks AI models to provide legal advice like a virtual lawyer or draft a judgment from the judge’s perspective, and then evaluates the answers. “If I don’t pass my knowledge to AI, others will do that,” Liu told Rest of World. She was aware AI could eliminate some jobs, she said, but saw the trend as unstoppable. For more stability, she was planning to take the civil service exam. From entertainment to productivity China’s data annotation companies have over the past decade recruited mostly low-paid workers, including student interns, rural residents, and stay-at-home mothers, to support the development of self-driving software and voice-recognition tools. The demand for data is now shifting away from consumer and entertainment applications toward productivity settings, intelligence firm IDC said in an August report. China’s AI training data market is expected to reach 7.8 billion yuan ($1.1 billion) in 2026, a 25% increase from 2025, according to IDC. Julie Yujie Chen, an associate professor at the University of Toronto who researches data labor in China, said middle-class workers were drawn to the data gigs due to the tough job market, and that government endorsement had lent more legitimacy to the profession. “The middle class in China with ordinary white-collar jobs are living in a state where they face this urgency to find a supplementary income,” Chen told Rest of World. Jessica, a Shanghai-based software engineer with two decades of experience, began taking AI training gigs in July. She had asked Claude about this new line of work, to which the chatbot responded that it was “the fastest way for ordinary people to get their hands on AI.” She completed tasks such as uploading text files to quiz AI on its ability to process lengthy texts and creating assignments to test AI’s agentic software skills. Jessica said the hundreds of yuan she earned was too little for the time she had spent. She also worried that the data platforms might misuse the work files she had uploaded. But she was willing to try: The software industry had suffered heavily from AI-driven disruption. At Jessica’s company, business slowed and full-time engineers were replaced by contractors. “You never know how long you will get to keep your job,” she said. “I want to have one more option. You have to keep evolving yourself to avoid getting completely replaced.”