How to label AI-generated content under Article 50 (with examples)

What Article 50 requires

Article 50 of the EU AI Act sets out transparency obligations for AI-generated content. The requirements differ by medium and by who is acting (provider vs deployer).

By medium

Text

AI-generated text published to inform the public on matters of public interest must be disclosed, unless a human exercises editorial review and takes responsibility for the content.

This is a scoped obligation — it does not apply to all AI-assisted text. Internal documents, marketing copy, and most business communications are generally outside this specific requirement.

Images, audio, and video

Providers of AI systems that generate synthetic audio, images, or video must ensure outputs are marked in a machine-readable way and detectable as artificially generated.

Deployers of deepfakes — AI-generated or manipulated content resembling real persons, places, or events — must disclose that the content is artificially generated or manipulated.

Emotion recognition and biometric categorization

Deployers of these systems must inform persons exposed to them.

Example labeling wording

AI-generated image

"This image was generated using artificial intelligence."

Place as a caption, alt text, or visible watermark.

Deepfake video

"This video has been artificially generated and may not depict real events."

Place at the start of the video and in the description.

AI-generated audio (podcast/voiceover)

"This audio was produced using AI voice synthesis."

Place in the audio introduction and in the content description.

Branch-specific limits and exceptions

• Personal, non-professional activity can fall outside the Act's scope
• The "obvious interaction" exception belongs to Article 50(1) AI interactions; it is not a general content-disclosure waiver
• Artistic, creative, satirical and fictional deepfakes still require an appropriate disclosure that does not hamper the work
• The public-interest text duty has an exception where a human exercises editorial control and a person holds editorial responsibility

Machine-readable marking

For providers, the obligation includes ensuring outputs are detectable as AI-generated. This may involve metadata, watermarks, or other technical markers. The Commission's code of practice on transparent AI, published in June 2026, provides further guidance.

What to do now

1. Audit what AI-generated content you publish
2. Determine which medium-specific rules apply
3. Apply appropriate labels at the point of publication
4. Log what was labeled, when, and by whom

The Labeling Kit includes a full guide, wording library, decision tree, and record log.