How do you label AI-generated content correctly?

Explicit and implicit labels

China's Measures for Labeling AI-Generated Synthetic Content define the two types clearly, and the same split is useful everywhere. An explicit label is added to the content or the interface and presented as text, sound or graphics that people can clearly perceive. An implicit label is added to the file data by technical means and is not easy for people to notice.

Neither type is enough alone. Visible labels inform viewers; machine-readable labels let platforms add their own notices automatically. Good practice is to use both and keep them intact when you edit.

China: the Measures for Labeling AI-Generated Synthetic Content

The Cyberspace Administration of China, with the Ministry of Industry and Information Technology, the Ministry of Public Security and the National Radio and Television Administration, issued the Measures in March 2025. They took effect on 1 September 2025. They apply to AI services already covered by China's rules on algorithm recommendation, deep synthesis and generative AI services.

• Explicit labels (Article 4): text gets a notice or symbol at the start, end or a suitable middle position, or in the interface; audio gets a voice or sound cue; images get a prominent mark in a suitable place; video gets a prominent mark on the opening frame and around the playback area; virtual scenes get one on the opening screen. Downloaded, copied or exported files must keep the explicit label.
• Implicit labels (Article 5): providers add metadata with the content's attributes, the provider's name or code, and a content number. Digital watermarks are encouraged.
• Platforms that distribute content (Article 6): if the metadata shows the content is AI-generated, the platform adds a prominent notice; if there is no metadata but the user declares it, the notice says it may be AI-generated; if neither, but the platform detects traces, it marks it as suspected AI-generated. Platforms must also offer a labeling function and remind users to declare.
• Unlabeled output (Article 9): a provider may give a user content without an explicit label only after the user agrees to take on the labeling responsibility, and must keep logs for at least six months.
• Users (Article 10): when you publish AI-generated content, you must declare it and use the platform's labeling function. Nobody may maliciously delete, alter, forge or hide the labels, or offer tools to do so.

Article 11 adds that labeling must also meet relevant laws, regulations and mandatory national standards, so providers need to follow the technical standards as well as the Measures.

European Union: AI Act Article 50

Article 50 of the EU AI Act (Regulation (EU) 2024/1689) sets transparency duties. Under Article 113 it applies from 2 August 2026.

• Chatbots (50(1)): people must be told they are interacting with an AI system, unless that is obvious.
• Machine-readable marking (50(2)): providers of systems that generate synthetic audio, image, video or text must mark the outputs in a machine-readable format so they can be detected as artificially generated or manipulated, as far as technically feasible. Assistive editing that does not substantially change the input is excluded.
• Emotion recognition and biometric categorisation (50(3)): deployers must inform the people exposed.
• Deepfakes (50(4)): deployers who publish image, audio or video deepfakes must disclose that the content is artificially generated or manipulated. For evidently artistic, creative, satirical or fictional work, the disclosure can be limited so it does not spoil the work. AI-generated text published to inform the public on matters of public interest must also be disclosed, unless a person or organisation holds editorial responsibility for it after human review.

A transitional rule in Article 111(4), added by a later amendment, gives providers whose systems were placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2).

Platform rules: YouTube, TikTok and Meta

Platforms apply their own rules worldwide, on top of local law. At a general level, at the time of writing:

YouTube does not require disclosure for clearly unrealistic content, minor edits such as color correction or filters, or AI help with scripts, titles and captions, and it says that disclosure alone does not limit a video's audience or monetization. TikTok says a label does not make harmful or misleading content acceptable.

How to label in practice

1. Keep the generator's metadata. Export from the AI tool in a format that keeps metadata, and check that your editor does not discard it on export.
2. Add a visible label that survives cropping. For video, put it on the opening frame and keep a small corner mark; for images, place it inside the picture, not only in the caption.
3. For audio-only content, say it at the start, for example: "This audio was generated with AI."
4. Turn on the platform's AI disclosure setting when you upload, even if the file already has metadata.
5. Write a plain caption line such as "Made with AI" or "AI-generated video" so viewers see it in every app.
6. Keep a record of the tools, prompts and source files for each published piece.

Be careful with privacy tools. Removing EXIF and GPS data, for example with cv.cm's exif tool, protects your location in photos you took yourself, but the same step also removes metadata labels from AI files. For AI-generated content, add a visible label before you share it, and do not remove labels that a law or platform requires.

Common mistakes

• Labeling only in the post text. Captions are lost when the file is reposted or downloaded.
• Placing a tiny label at the very edge where platform buttons or cropping remove it.
• Converting or screen-recording an AI video, which removes its metadata, and then publishing it without a visible label.
• Thinking a cartoon style needs no label everywhere. Western platforms focus on realistic content, but China's Measures apply to AI-generated content in general.
• Labeling real footage as AI to avoid responsibility. China's Measures forbid using improper labels to harm the rights of others, and a label never makes misleading content acceptable.