By August 2026, AI content labelling had moved from rulemaking into ongoing implementation and targeted enforcement. Before publishing video containing generated or synthetic images, sound or text, businesses should check AI use, visible labels, metadata, platform declarations, rights, accuracy and review records.
This checklist helps brands, manufacturers, advertisers, marketing departments and video project owners reduce missed labels and publishing errors. It does not replace current platform checks or project-specific legal review.
1. Why has labelling become an operational issue?
The Measures for Labelling AI-Generated Synthetic Content and mandatory standard GB 45438–2025 took effect on 1 September 2025. They distinguish explicit labels, perceptible as text, sound or graphics, from implicit labels embedded in file data with attributes and production information.
The Clean and Clear campaign against misuse of AI applications, launched in April 2026, included inadequate labelling among its priorities. The CAC's 6 July update covered registration, security reviews, data poisoning, false information and labelling. Overall enforcement totals cannot all be attributed to labels, but labelling is clearly more than an optional internal note.
The CAC reporting centre also opened a dedicated AI-misuse reporting section in June 2026, including inadequate labelling. Businesses should record AI use throughout planning, production, delivery and upload, rather than deciding whether to mention it only after export.
2. Identify generated and synthetic content
The measures cover AI-generated or synthetic text, images, audio, video and virtual scenes. Check not only fully generated films but backgrounds, synthetic people, digital humans, voice cloning, AI narration, product scenes, replacement shots and generated caption copy.
Maintain an asset list recording names, suppliers, generation or filming methods, tools, permissions, changes and final placement. This also supports checks of product appearance, personal permissions, fonts, music and commercial-use scope.
3. Six pre-publication checks
First, inspect visible labels. For video within the relevant requirements, providers should add prominent indications at the beginning and in suitable playback-adjacent positions. Ensure labels remain legible at actual publication sizes and survive cropping, introductions and re-encoding.
Second, inspect implicit file labels. Relevant providers must embed attributes, provider names or codes and content identifiers in metadata. Keep original exports and check metadata through transcoding, subtitling, templating and upload.
Third, use the platform's declaration tools. Publishers of generated or synthetic content should declare it and use platform labelling functions. An on-screen notice does not replace the upload-page declaration. Check the current interface for each platform and content type on publication day.
Fourth, do not maliciously remove, alter, forge or conceal labels. Include labelling in checks for re-edits, resizes and excerpts. Removing ordinary watermarks must not be confused with deleting legally required AI labels.
Fifth, verify people, products and facts. An AI label does not establish truthfulness, rights clearance or advertising compliance. Review likenesses, voices, trademarks, fonts, music, specifications, efficacy claims and scene authenticity separately.
Sixth, retain publication evidence: final files, asset and permission lists, review comments, versions, declaration screens and published screenshots. Assign a final approver across channels to avoid inconsistent label states.
4. Build checks into production
Mark AI shots during scripting or storyboarding, retain tool and asset records during production, check rights and facts at first review, inspect labels and cropping at final review, declare AI use during upload and archive the published page.
Check every horizontal advertisement, vertical short, silent-screen and sales version separately. A compliant master does not guarantee that cropped, compressed or re-edited files retain the same labels and metadata.
5. Three common misconceptions
A note saying “made with AI” in the description is not necessarily enough. Descriptions, visible labels, metadata and platform declarations have distinct roles and are not simple substitutes.
Automatic platform detection does not replace the publisher's declaration. Detection is part of platform governance, not a reason to skip publication checks.
An AI label does not clear likeness, voice, copyright, trademark, personal-information, trade-secret or misleading-advertising issues. Rights and accuracy still need separate review.
6. FAQ
If AI only assists storyboarding and the final footage is entirely live action, does the finished film need a label? Assess the final public content, actual AI use, provider terms and platform requirements. Internal storyboards are distinct from public outputs; retain records rather than treating every AI-assisted task identically.
Must a few seconds of AI imagery be checked? Yes. Duration alone is not decisive. Identify how the clip was generated and where it will appear, then apply the relevant rules and platform requirements.
What if the AI tool's export already has labels? Check that later editing has not damaged labels or metadata, make platform declarations and separately verify permissions and support for people, voices, music, fonts, product information and advertising claims.
Commercial and product video production: https://raystudio.art/services/dongguan-guanggaopian
Related reading on AI-video contracts, rights and labels: https://raystudio.art/news/yzc9kmfp4fcddwktl34i3x5f
AI video, commercial or product project brief: https://raystudio.art/contact#project-brief-form
7. References and content note
CAC, first-stage update on the campaign against AI application misuse, 6 July 2026: https://www.cac.gov.cn/2026-07/06/c_1785081384384987.htm
CAC, dedicated AI-misuse reporting section, 12 June 2026: https://www.cac.gov.cn/2026-06/12/c_1782660002371256.htm
CAC and three other departments, Measures for Labelling AI-Generated Synthetic Content, published 14 March 2025, effective 1 September 2025: https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm
SAMR and SAC, GB 45438–2025, published 28 February 2025, effective 1 September 2025: https://openstd.samr.gov.cn/bzgk/std/newGbInfo?hcno=F32EA2A561F1886CD8D606513512D547
The Chinese article was checked on 24 August 2026. Prepared by Guangdong RAY STUDIO from public rules and publishing workflows; not legal advice. Advertising, digital humans, voice cloning, news, film production and cross-border publishing require separate review against applicable rules and platform requirements.
