Do I have to label AI content?
Reviewed
Short answer
Only in specific cases: under Article 50 of the EU AI Act, anyone using AI professionally must visibly disclose deepfakes (realistic image, audio or video that resembles real people, places or events and could pass as authentic) and AI-generated text published to inform the public on matters of public interest, unless a human reviewed it under someone’s editorial responsibility. Separately, the company that provides the AI tool must embed a machine-readable mark in every output, and purely personal, non-professional use is outside the Act.
Official sources: AI Act, Article 50 (consolidated) · Commission guidelines on Article 50
Who has to do what
| You are | What Article 50 asks of you | From |
|---|---|---|
| A company that offers an AI generator (provider) | Mark every output in a machine-readable format so it is detectable as AI-generated (50(2)); tell people when they interact with an AI system (50(1)). | 2 August 2026; generators already on the market before then: 2 December 2026 for marking |
| A business, agency, freelancer or creator using AI for work (deployer) | Visibly disclose deepfakes; label AI-generated text published on matters of public interest unless it was human-reviewed under editorial responsibility (50(4)); inform people exposed to emotion recognition or biometric categorisation (50(3)). | 2 August 2026 |
| A private person, purely personal and non-professional | No duty under the AI Act (Article 2(10)). The tool’s provider still has to mark the output. | — |
| Anyone, for content published before 2 August 2026 | No retroactive labelling required; the Commission encourages it where possible. | — |
“Professional” is broad. The Commission’s guidelines treat any activity through which a person gains an economic benefit on a regular basis, or that is part of a business, trade, occupational or freelance activity, as professional. Their examples of personal use include generating deep fakes of members of your household, a student writing homework with AI, and an individual posting deep fakes on social media in a personal capacity. A creator who earns from their channel regularly is therefore a deployer.
Does your content need a visible label? Three questions
- Do you use the AI tool for work, or is your use purely personal and non-professional? If purely personal, the AI Act does not apply to you (platform rules may still).
- Is it image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful? Then it is a deepfake (Article 3(60)) and needs a clear disclosure. Is it text published to inform the public on matters of public interest, with no human review under editorial responsibility? Then it needs a label too.
- Is the deepfake part of an evidently artistic, creative, satirical or fictional work? Then the duty is lighter: disclose in an appropriate way that does not hamper the display or enjoyment of the work.
The label must reach people “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure” and meet accessibility requirements (Article 50(5)). The guidelines add that deployers cannot rely on the invisible, machine-readable mark the provider embedded: the disclosure has to be perceivable without tools, for example a visible or audible label.
By type of content
| Content | Needs a disclosure (examples) | Usually does not (examples) | Guide |
|---|---|---|---|
| Images | A manipulated image of two real footballers in front of a stadium-like building; a product image that makes the product look better than it is | A sphinx flying over the Eiffel Tower; colour correction | AI images |
| Video | A realistic avatar of a company CEO congratulating staff; an AI speech by someone resembling a politician | Mice arguing about cheese in an advert; AI backgrounds behind real actors | AI video |
| Voice and audio | A cloned voice of a podcast’s regular presenters discussing the news | Voices for clearly fictional characters in audiobooks or games; noise reduction | AI voice and audio |
| Music | AI music resembling the style of existing artists (a deepfake inside an artistic work: lighter disclosure) | — | AI music |
| Text | AI articles on EU policy posted without any human review | A newspaper article reviewed by its editor-in-chief; internal documents | AI text |
| Avatars and deepfakes | A realistic synthetic influencer testing a sponsored product | AI gaming worlds such as fictitious forests and castles | Deepfakes and avatars |
Platforms ask for more
Platform rules are separate from the law and often stricter. YouTube requires creators to disclose realistic AI-generated or meaningfully altered content and AI music that is the main focus of a video. TikTok requires creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people. Meta requires its disclosure tool for organic posts with photorealistic video or realistic-sounding audio that was digitally created or altered. Each type guide summarises the rules that matter for that kind of content.
How to label, in practice
- Put a visible (or, for audio, audible) disclosure on the content itself, seen at the first exposure, not only in the terms of use.
- Use the optional EU icons or an equivalent plain-language label such as “AI-generated”.
- Keep the machine-readable marks the AI tool embedded; re-exports and screenshots can remove metadata.
- Tick the platform’s own AI setting when you upload (YouTube, TikTok, Meta, Pinterest).
- Keep a short record of which assets were made or changed with AI and how you labelled them.
Questions
- I post AI images on my personal account. Do I have to label them?
- Not under the AI Act, if the use is purely personal and non-professional: Article 2(10) excludes it, and the Commission’s guidelines give personal deep fakes shared on social media as an example. If you earn from the account on a regular basis, the guidelines treat it as professional use. Platform rules can still require a label.
- Is a C2PA tag or an invisible watermark enough as a label?
- No, not for the deployer’s duty. The Commission’s guidelines say deployers cannot rely on the machine-readable marking embedded by the provider, because it is not immediately clear to people; the disclosure must be perceivable without tools. The machine-readable mark is the provider’s separate duty under Article 50(2).
- Do I have to label content I published before 2 August 2026?
- No. The Commission’s questions and answers on Article 50 say content generated before 2 August 2026 does not need to be labelled retroactively, although it encourages deployers to do so where possible.
Sources
- AI Act, Article 50: Transparency obligationsEuropean Commission, AI Act Service Desk (AI Act Explorer)Consolidated text as at 27 July 2026 (includes the Digital Omnibus, Regulation (EU) 2026/1744), shown in the Commission’s AI Act Explorer.read
- AI Act, Article 2: ScopeEuropean Commission, AI Act Service Desk (AI Act Explorer)Consolidated text as at 27 July 2026 (includes the Digital Omnibus, Regulation (EU) 2026/1744), shown in the Commission’s AI Act Explorer.read
- Guidelines on the transparency obligations under Article 50 of the AI Act (C(2026) 5054, 20 July 2026)European Commissionread
- Transparency obligations under Article 50 of the AI Act — questions and answersEuropean Commissionread
- Quick Facts: Transparency rules for AI systemsEuropean Commissionread
- EU Icons for labelling AI-generated contentEuropean Commissionread
- Disclosing use of GenAI contentYouTube Helpread
- Community Guidelines: Integrity and Authenticity — Edited Media and AI-Generated Content (AIGC)TikTokread
- Labeling AI-Generated Images on Facebook, Instagram and ThreadsMeta Newsroomread
Last reviewed by Volkan Çatak, Founder, Proofcan.
Not legal advice. This page explains official texts in plain words and links each one; check the sources, or a lawyer, before relying on it for a decision.