Labeling of AI-generated and synthesized content
It has to be judged in two cases. Where an enterprise itself provides AI-generated and synthesized services to others (for example an externally launched AI customer service, AI writing tool or AI question-and-answer bot), it is a service provider regulated by the Measures and should, under Articles 4, 5 and 8, implement the explicit label, the implicit label in file metadata, and the explanation of labeling method and styling in the user service agreement. Where an enterprise only uses AI as an internal tool to draft copy and then posts the content on a platform, the Measures directly regulate service providers and dissemination platforms, but at the posting stage Article 10 requires the user to declare it voluntarily and to use the labeling function provided by the platform.
Article 2 defines the regulated subjects: “These Measures apply to network information service providers that meet the situations set out in the Provisions on the Administration of Algorithmic Recommendations in Internet Information Services, the Provisions on the Administration of Deep Synthesis of Internet Information Services, and the Interim Measures for the Administration of Generative AI Services (hereinafter service providers) when they carry out labeling activities for AI-generated and synthesized content.”
Article 3 defines the content: “AI-generated and synthesized content means information such as text, images, audio, video and virtual scenes generated or synthesized using AI technology.” Marketing copy and customer service replies take the form of text and fall within the enumerated scope.
Article 4 places the explicit label requirement on the service provider: “Where the generated and synthesized service provided by a service provider falls within the situations in Article 17, paragraph 1 of the Provisions on the Administration of Deep Synthesis of Internet Information Services, it shall add an explicit label to the generated and synthesized content in accordance with the following requirements:” for text content, a prominent prompt may be added in the interactive scenario interface or around the text, which gives AI customer service a clear path for its interface prompt.
Step one is to decide your identity: are you a “service provider offering generated and synthesized services to others” or a “user of someone else’s service”. The test is whether the service form falls within the situations listed in Article 2 of the Measures, not whether you trained the model yourself.
If you provide the service: implement the explicit label under Article 4 (AI customer service can prompt in the interactive scenario interface), write the implicit label into file metadata under Article 5, explain the labeling method and styling in the user service agreement under Article 8, and prepare filing and security assessment materials as required.
If it is internal use only, published on your own channels only: declare it voluntarily under Article 10 when the content enters a dissemination platform and use the platform’s labeling function; for your own channels such as official accounts and your website, it is advisable to state AI involvement in the content to reduce the risk of misidentification.
AI customer service: give a prominent prompt in a fixed position of the dialogue interface, not only a one-off pop-up in the first session.
Marketing copy: keep the manual review records and content provenance, and avoid publishing undeclared, especially for material featuring a person’s image, a celebrity voice or a fictional scene.
Skipping every duty with the line “we only write copy with AI, we are not a service provider”, and ignoring the user-side declaration duty at the posting stage.
Assuming AI customer service replies are too short, or are dialogue rather than content — text is one of the forms explicitly listed in Article 3.
Building an externally facing AI tool without explaining the labeling method and styling in the user agreement (Article 8).
Pushing all labeling responsibility onto the large-model vendor used, and ignoring the fact that you provide the service to users.
Adding a prompt only to the final product, so exported, downloaded or copied files carry no label.
Q: Internal knowledge base question and answer, not offered externally — does it need a label?
A: The Measures apply to qualifying network information service providers (Article 2), so whether an internal system falls inside depends on the service form; the local cyberspace administration office sets the operative reading on boundary cases.
Q: When generating marketing copy through a large-model API, who is responsible for the label?
A: The API provider has its own duties as a service provider; and when the enterprise posts the content on a dissemination platform, the Article 10 duty to declare voluntarily rests with the enterprise. The two do not replace each other.
Q: Must AI customer service always carry a prompt in the interface?
A: Article 4 allows text content to carry a prominent prompt in the interactive scenario interface or around the text. AI customer service is an interactive scenario, so a prominent prompt in the interface is a workable and prudent approach.
This entry belongs to the “China AI Compliance and Filing” cluster (42 entries in total): View all · Answers home
Who we are: a knowledge base and AI customer service team serving small and micro businesses, delivering implementations that take AI customer service through the national standard and the filing process.
Content on this site is compiled from publicly released regulatory texts for enterprises’ own self-check reference; the specific filing requirements of the local cyberspace administration authorities prevail, and this is not legal advice.
This page is generated from the China AI compliance and filing MCP corpus — agents can call the same data directly.