A viewer lands on a livestream selling skincare. The host is polished, never tires, and answers every question at midnight. Half the time, nobody told the viewer that the host is not a person at all.
That gap — between what looks human and what is machine — is exactly what a stack of new Chinese rules is trying to close.
The national labeling floor
The foundation is 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 together with the Ministry of Industry and Information Technology, the Ministry of Public Security and the National Radio and Television Administration, and effective 1 September 2025.
Its logic is simple: anything an AI makes — text, images, audio, video, or virtual scenes — must carry both:
- An explicit label a user can actually see (a tag or prompt on the content), and
- An implicit label buried in the file's metadata, so the origin stays traceable as the file travels.
Crucially, the measures pull distribution platforms into the loop, not just the tools that generate the content. If a platform detects AI痕迹, it has to add a visible warning at the edges of the post.
The livestream-specific rule
For shopping livestreams, the State Administration for Market Regulation floated dedicated Measures for the Supervision and Administration of Live-Streaming E-Commerce in a June 2025 draft. The relevant clause — Article 30 — says a live room using AI-generated person images or video for marketing must display a prominent, persistent label telling viewers the figure is AI-generated, clearly set apart from a real human host.
The draft also bans using AI to fabricate or spread false information or to impersonate others for marketing, and it introduces a platform "blacklist" system: serious offenders get reported to provincial regulators and, ideally, shared across platforms.
Eight platforms sign on
The most concrete industry move came from the Beijing Consumer Association, which in early December 2025 got eight e-commerce platforms — JD, Meituan, Pinduoduo, Vipshop, Douyin, Kuaishou, Xiaohongshu and WeChat Live — to sign the country's first Commitment on Promoting the Standardized Application of AI Technology.
The pledge's demands:
- Merchants must clearly mark AI-generated or "simulated-scene" content at a prominent spot.
- AI face-swap impersonation of celebrities for selling is explicitly prohibited.
- Platforms must monitor livestreams and, if they find an unlabeled AI digital human (数字人) or virtual scene, add a warning and, where it breaks the rules, shield the link.
- Deleting, forging or hiding AI labels is treated as a violation to be punished.
The association's own survey is what justified the urgency: of ten AI digital-human livestream rooms it checked, half carried no "AI-generated" or "simulated-scene" tag at all. Nearly 30% of consumers said they worried AI-generated content could mislead their purchase decisions.
Why this is consumer protection, not censorship
The framing here is narrow and practical: a buyer has a right to know whether the persuasive face on screen is a person with opinions or a rendered avatar with a script. The rules do not ban digital humans — China's livestream commerce leans heavily on them for round-the-clock selling — they just require the curtain to be labelled.
That distinction matters. The measures sit inside a broader, already-operating labeling framework rather than a sudden new restriction, and the Beijing pledge is voluntary industry self-discipline backed by consumer protection, not a criminal statute.
What readers can do now
- If you shop livestreams in China, look for the "AI生成" or "拟真场景" tag before trusting a host's claims; if a digital human is selling and the label is missing, that itself is a red flag worth reporting to the platform.
- If you run a brand or store, bake the explicit label into every AI-hosted stream from the first frame — the rules make the room operator, not the tech vendor, liable for unlabeled AI content.
- If you build AIGC tools, treat dual labeling (visible tag plus metadata watermark) as a default feature to ship, because platforms are now mandated to check for it on upload.
Honest limitations
Core facts (the AI-content labeling Measures issued by CAC, MIIT, MPS and NRTA, effective 1 September 2025, requiring explicit and implicit labels across text/image/audio/video/virtual scenes and pulling distribution platforms into compliance; SAMR's June 2025 draft live-streaming e-commerce measures with Article 30 mandating a prominent persistent label for AI-generated person images/video and a platform blacklist; the Beijing Consumer Association's December 2025 pledge signed by JD, Meituan, Pinduoduo, Vipshop, Douyin, Kuaishou, Xiaohongshu and WeChat Live, banning AI face-swap celebrity impersonation and requiring link-shielding of unlabeled AI content; the association's survey of 10 digital-human rooms with half unlabeled and ~30% of consumers worried about AI misleading purchases) are sourced from the CAC official site (cac.gov.cn), China's central government portal (gov.cn), Xinhua, China National Radio, People's Daily Online (people.com.cn) and Workers' Daily (workercn.cn), 2025. The SAMR e-commerce measures were a draft for public comment as of mid-2025, not a finalized binding statute at that stage; the Beijing pledge is a voluntary industry commitment, and enforcement of labeling ultimately depends on platform implementation. No RMB figures appear in this article, so no currency conversion applies. Analysis is current to 29 September 2026.