A labelling rule exists before a usable AI share does

The figure or the record

China's Cyberspace Administration issued a notice on labelling AI-generated content, published 14 March 2025 and effective from 1 September 2025, requiring explicit and implicit labels on AI-generated text, audio, images and video. The National Radio and Television Administration's Order No. 16 on micro-drama development, published 31 July 2026 and effective from 1 September 2026, applies that requirement specifically to micro-dramas: Article 34 states that a micro-drama 'generated or produced using artificial intelligence technology' must carry a prominent notice on each episode. Separately, the China Netcasting Services Association's main findings from its 2025 industry white paper, published 11 November 2025, state that more than 60% of surveyed practitioners named AI and other new technology as an important driver of continued growth, and that unnamed research shows roughly a 50% average efficiency gain from adopting AIGC tools.

What the documents establish

The two regulatory documents establish a labelling obligation, which is the mechanism that would make a real count of AI-generated releases possible in principle, once episodes carry the required marks. The white paper establishes something different: a practitioner-survey figure about the perceived importance of AI, not a count of how many released titles actually used it, and an efficiency claim attributed only to unnamed research rather than a named, checkable study.

Scope and caveats

A labelling requirement effective from September 2025 or September 2026 does not retroactively mark earlier releases, and neither regulatory document states what share of current releases carries the label. The white paper's figure of more than 60% measures opinion among survey respondents, not the proportion of titles produced with AI tools, and the two should not be substituted for each other.

What to watch

An editorial framework for any claim that most releases used AI: identify whether it is a labelled count, a survey of opinion, or an unattributed estimate, and treat only the first as a genuine share of releases.

• Does the claim come from a count of labelled content, a survey of practitioners, or an unnamed study?
• Does the source name a specific sample size, period and question wording for any survey figure?
• Would the labelling rules in force at the time have actually required disclosure for the content being counted?

A rule that requires a label is a precondition for a trustworthy share claim, not a substitute for one, and the white paper's own figures show the difference between the two.