AI content labeling

What must platforms verify and flag on AI content?

What you may need to do

1. Build metadata verification capability that separates the three cases and handles each one differently
2. Design different notice wording for the three cases, rather than describing everything as AI-generated
3. Provide an entry point on the publishing page for declaring whether the content contains generated or synthesized content
4. For cases (1) to (3), add the dissemination element information into the file metadata

Clause basis

One-line answer

A platform providing online information content dissemination services (short video, social, content communities and the like) must handle the cases separately under Article 6: where verified metadata clearly marks the content as generated or synthesized, put a prominent notice around the published content and clearly tell the public that the content is generated or synthesized; where no implicit label is verified but the user declares the content to be generated or synthesized, notify the public that the content may be generated or synthesized; where neither is present but the platform detects an explicit label or other traces of generation or synthesis, identify the content as suspected generated or synthesized content and say so. The platform must also provide labeling functions and remind users to declare voluntarily.

The basis

Article 6 opens with: “Service providers that provide online information content dissemination services shall take the following measures to regulate the dissemination of generated and synthesized content”.

Item (1): “verify whether the file metadata contains an implicit label; where the file metadata clearly marks the content as generated or synthesized content, take appropriate means to add a prominent notice around the published content, clearly reminding the public that the content is generated or synthesized content”.

Item (2): “where no implicit label is verified in the file metadata, but the user declares the content to be generated or synthesized content, take appropriate means to add a prominent notice around the published content, reminding the public that the content may be generated or synthesized content”.

Item (3): “where no implicit label is verified in the file metadata and the user has not declared the content to be generated or synthesized content, but the service provider providing online information content dissemination services detects an explicit label or other traces of generation or synthesis, identify it as suspected generated or synthesized content, take appropriate means to add a prominent notice around the published content, reminding the public that the content is suspected to be generated or synthesized content”.

Item (4): “provide the necessary labeling functions, and remind users to declare voluntarily whether the content they publish contains generated or synthesized content”.

In addition, in the cases described in items (1) to (3), the platform shall also add dissemination element information to the file metadata: the attribute of the generated or synthesized content, the name or code of the dissemination platform, and the content number.

What you need to do

Build metadata verification capability that tells apart three situations: verified, not verified but declared by the user, and neither present but traces detected.

Design different notice wording and display for the three situations, so that suspected content and confirmed content are not described in the same sentence.

Provide a declaration entry point inside the publishing flow and prompt the user actively, rather than hiding it in settings.

For cases (1) to (3), add the dissemination element information to the file metadata (content attribute, platform name or code, content number).

Keep a record of how verification and notices were handled, so that you can explain the case to the competent department afterwards.

Common pitfalls

Building only the publisher-side declaration entry point and doing no metadata verification, so that the first case cannot be identified at all.

Using one sentence — “this content is AI-generated” — for all three cases, which does not match the differentiated wording the article requires.

Hiding the notice on a second-level page or at the bottom of a detail page, where the public cannot see it in the feed.

Only displaying the user declaration in the front end, with no record kept, so that nothing can be evidenced afterwards.

Overlooking the platform-side duty to add the dissemination element information to the metadata.

Common follow-up questions

Q: If the platform has added a notice, does the author still need to declare anything?

A: The two do not conflict. Article 10 requires the user to declare voluntarily and to use the labeling function the platform provides; the platform-side duty and the user-side duty run in parallel.

Q: If traces of generation or synthesis are detected, does that confirm the content is definitely AI-generated?

A: For this situation the article says the content is identified as “suspected generated or synthesized content”, and the wording of the notice should reflect that same “suspected” quality.

Common follow-up questions

If there is only a user declaration and no metadata, must the platform still give a notice?

Yes. Article 6, item (2) covers this situation: notify the public that the content may be generated or synthesized content.

Where does a notice count as being “around the published content”?

The article says “take appropriate means to add a prominent notice around the published content”. The exact position is judged by whether the public can clearly see it while browsing.

This entry belongs to the “China AI Compliance and Filing” cluster (42 entries in total): View all · Answers home