China's AI course boom sells hope — and Beijing calls it leek-cutting
China is answering the AI shock the way it has answered every shock for forty years — enroll in something — and a paid-course economy has grown up around the dread. The state's own press is now calling it a fleecing. Also: OpenAI will target ads with contact details you never handed it.
Millions of Chinese professionals and students are paying for AI courses sold on the fear of being left behind, and Chinese state media has started describing the market as a racket that "takes advantage of people's 'skill anxiety' about new technologies." New York Magazine's Afra Wang spent months reporting the phenomenon from Hangzhou and the rust-belt northeast: self-organized classes in living rooms where the first lesson is teaching newcomers to "put up the ladder" — the euphemism for the VPN relays that get around OpenAI's geo-block — sitting alongside a formal industry of paid memberships, offline bootcamps and co-working floors. The reflex is old. During reform-era "English fever," fluency was sold as the ticket out; the conviction has simply changed languages.
The commercial scale is the part that surprises. One of the largest "make money with AI" operators, Shengcai Youshu — roughly "The Art of Getting Rich" — charges ¥3,365 a year for community access against an average Hangzhou private-sector salary of ¥7,671 a month: half a month's wages. The company says it has taken more than 80,000 paying members in nine years and runs over 2,000 offline events a year, with gamified tiers, points called Dragon Balls, and a pitch delivered in the register of a revival meeting — AI as your "silicon-based labor force" that works 7-by-24.
Beijing's own press has stopped defending the category. Economic Daily, carried on People's Daily's site, documented the speed-course advertising stack — "run the whole AI pipeline in 90 minutes," "learn AI monetization in three days," "guaranteed ¥8,000 a month after graduation" — and reported that the China Consumers Association logged average claims above ¥7,000 in vocational-skills complaints in the first quarter of 2026, with digital-skills courses singled out for false advertising and "learn now, pay later" loan traps. Xinhua's reporting earlier this year described a grey supply chain aimed at students; Sichuan's consumer council counted 3,739 "AI+" training complaints in 2025.
What makes this more than a consumer-protection story is who is buying. Wang found people who cannot afford to be wrong: a security guard who spent six months with Claude Code on his phone and is about to start at a semiconductor-packaging company, and a 25-year-old from the northeast who taught himself AI filmmaking and now runs a studio turning over several million yuan a year, cutting three-minute government documentaries for around ¥100,000 that would have cost over ¥1 million shot conventionally. The same market that fleeces also, sometimes, delivers — which is precisely why it keeps selling. Beijing wants the upskilling and is funding its own version of it: a five-ministry "AI + education" action plan in April, a five-million-person talent gap cited by the human resources ministry, and subsidized training schemes in Zhejiang, Jiangsu and Shenzhen. It just wants the courses to stop being a racket.
Advertisers can now target you inside ChatGPT with a list of email addresses and phone numbers they already hold — and OpenAI's ad platform accepts them raw, not just hashed. 404 Media's Joseph Cox bought his own campaigns on the self-serve system and documented the mechanics: an advertiser supplies raw or hashed emails, phone numbers, or Android advertising IDs, OpenAI matches them against its user base, and the matches can then be included in or excluded from a campaign. Lists under 25,000 entries only work for exclusion, which quietly tells you the matching happens at population scale rather than one name at a time. An OpenAI spokesperson told 404 Media the uploaded files are deleted after processing, usually within 24 hours, and that advertisers cannot see which individuals matched or pick specific people out of an audience.
The prompt side has no real precedent. OpenAI lets advertisers write "context hints" describing the conversations, topics and keywords their product should sit beside, and Cox proved the lever works by buying ads that read "ChatGPT isn't private. OpenAI hires humans to read users' chats" against privacy-related prompts. The personalization opt-out is narrower than its wording suggests — ChatGPT's own setting says turning it off does not stop ads, and that ads "may still be based on your current chat." For a product with a billion monthly active users, most of whom type things they would not post on a social network, that is the substantive change: the chat is now a targeting signal, and the target list can be assembled before the user ever opens the app. This is ordinary adtech landing somewhere new, and the people who measure adtech say so. "They are boiling the frog and hoping their users won't notice the privacy implications," DecryptAds co-founder Zach Edwards told 404 Media, arguing that advertisers handing over sensitive personal data should have to register as data brokers. We mapped the other half of this system in September — OpenAI's ad pixel tracks ChatGPT users across 1,000 websites — and the two mechanisms now cover both directions: what you type in, and what someone else already knows about you.
What to watch: whether any regulator treats uploaded contact lists as data-broker activity, and whether China's next consumer-protection round turns into enforcement rather than commentary.
Would you pay half a month's salary for an AI course if your job depended on it? Tell us in the comments.
