Liability sits with brands and platforms, because the Advertising Law treats fabricated product effects as false advertising, and lawyers say an AI label does not cure a false claim.
An advertising planner told Changan Street Knower (长安街知事), the new-media outlet of Beijing Daily, that a traditional 60-second feed ad shot in Beijing by a three-person crew costs at least RMB10,000 and takes a week or more. With AI, she said, a skilled operator with a strong computer can produce dozens of videos with different scripts, personas and backgrounds in one evening, for a few hundred yuan of electricity and software subscriptions. The report, carried by Sina Finance on October 18, describes a growing trade in ads that show striking before-and-after results and polished "elite" figures vouching for products, many of them generated rather than filmed.
China’s rules already require AI content to be labeled and bar endorsers from vouching for products they have not used. What the report adds is a picture of the supply side: agencies selling cheap tools that promise output platforms cannot spot, while the cost of checking each ad stays with reviewers, platforms and regulators.
For brands, the consequences can arrive quickly. Supor’s e-commerce accounts began posting batches of AI-generated short videos in late July built on suggestive scenes, and The Beijing News noted in an August 9 commentary that the company’s shares had fallen four days in a row as criticism spread.
A few tools and a few hundred yuan now replace a film crew
The planner, identified only as Ms. Hu, described her workflow to the reporter. She has ChatGPT write 100 ad scripts and keeps 10, builds actors and sets in Midjourney, animates the images with Runway and adds a synthetic voice. Each product can then be pitched in separate versions to young mothers, office workers and older viewers, the last shown messages about long, healthy lives.
Liang Yiran, media manager at Shijiazhuang Angyu Electronic Technology, offered a more qualified view. High-quality AI work for large brands is far from cheap, Liang said, and at equal cost filmed footage usually looks better. The appeal is strongest for small merchants and marketing accounts, for whom low cost matters more than polish.
Posing as a merchant, the reporter contacted three AI marketing firms. At one, AI Jixiang Technology, a salesperson offered a talking-head avatar tool for RMB899 with two years of free computing and said its videos would carry no AI label. The salesperson also claimed platforms could not detect the content. The other two quoted several hundred to several thousand yuan for full-service AI marketing. The Beijing News commentary put the gap starkly: an AI ad can cost a few yuan, while paying a person to review it costs tens of yuan.
The labeling rules address disclosure and leave false claims to advertising law
The Measures for Labeling AI-Generated Synthetic Content (《人工智能生成合成内容标识办法》), issued by the Cyberspace Administration of China with three other agencies and in force since September 1, 2025, require visible labels and embedded metadata on AI-generated video, images, audio and text. Platforms must check for the embedded markers, and maliciously removing or hiding a label is banned. The 899-yuan offer of unlabeled output runs directly against them.
Labels govern disclosure only. Dong Shilian, director of Beijing Fengtai Law Firm, told the reporter that labeling serves consumers’ right to know but leaves the standard of truthfulness unchanged; a false ad stays false with a label on it. A lawyer quoted by Xinhua in June made the same point about sellers who add an "AI生成" tag to an exaggerated product video.
That standard sits in the Advertising Law (《中华人民共和国广告法》), which the market regulator administers. It counts fabricated product effects, and invented or unverifiable statistics and quotations offered as proof, as false advertising. It also requires endorsers to speak from fact and bars them from vouching for anything they have not used. Endorsers who vouch for products they have not used face confiscation of illegal gains and fines of one to two times those gains. Ads for medicines, medical devices and health foods may not use endorsers at all.
Regulators are starting to treat avatars as the seller’s responsibility
On September 7 the State Administration for Market Regulation (SAMR) released seven cases of false advertising in online food sales, China News Service reported. In the first, Beijing Shuangtai Yuan Trading used an AI virtual human in a livestream to claim an ordinary food boosted immunity; the Changping district regulator ordered it to stop and fined it RMB5,000 under the Anti-Unfair Competition Law (《中华人民共和国反不正当竞争法》) and the Livestream E-Commerce Supervision Measures (《直播电商监督管理办法》). The case summary says an AI avatar is no shield against a false-advertising finding.
The Beijing Municipal Administration for Market Regulation (北京市市场监督管理局) followed on October 8 with a consumer warning, reported by Southern Metropolis Daily, that a virtual figure does not count as a real person’s endorsement and that unlabeled AI ads are likely noncompliant. It added that sales figures, reviews, test reports and expert quotes in ads may themselves be AI inventions.
Dong argues that the burden should shift to platforms at the source, with separate review channels for filmed and AI-generated material and a requirement that merchants declare which kind they are submitting. For high-risk categories, including medical aesthetics, food, health products, medical devices and education, he wants a human to sign off on every efficacy claim, persona and scene. A major short-video platform adopting a declared-material review track would be the clearest sign of change.