How to add AI disclosures that comply with the EU AI Act
AI-generated content disclosures become mandatory under EU AI Act Article 50 from August 2026. This guide covers where, when, and how to add compliant labels to your content — including the EU's official AI-content icons.
The EU AI Act’s transparency obligations under Article 50 apply from 2 August 2026. If you publish AI-generated content professionally, you have specific disclosure duties. These duties go well beyond a footnote in your terms and conditions. The European Commission published draft guidelines that explain what disclosures must look like, where they must appear, and what does not meet the standard. This guide translates those requirements into practical placement guidance for content creators, freelancers, and science communicators.
Note: The guidelines referenced throughout this post are a draft published for stakeholder consultation. They do not prejudge the final version, which will be adopted by the Commission. The legal source is Article 50 of Regulation (EU) 2024/1689 (the AI Act).
Does Article 50(4) of the EU AI Act apply to your content?
Article 50(4) creates two separate obligations, depending on what you publish. The first applies to deep fakes: AI-generated or manipulated image, audio, or video content that resembles real persons, objects, places, or events, and would falsely appear authentic or truthful to a person. The second applies to AI-generated or manipulated text published to inform the public on matters of public interest.
For science communicators and content creators, the text obligation is the most relevant one. The Guidelines describe “matters of public interest” as topics relevant to society at large. These include public health, environmental protection, consumer safety, and scientific, political, or cultural developments with potentially important public implications. The list is not exhaustive.
The Guidelines give examples of in-scope content: an AI-generated summary of a newspaper article on a town council decision, an AI-manipulated academic paper that compares the effects of diets on a disease, and an AI-generated message on a meteorological institute’s social media profile that warns citizens about stormy weather. Out-of-scope content includes AI-generated fictional novels or poems, AI-manipulated advertisement text (unless it contains health, safety, or sustainability claims), and AI news summaries that a chatbot delivers only to the user who prompted it.
If your content explains research findings, covers public health, reports on policy, or addresses environmental topics for a public audience, it almost certainly falls within the scope of Article 50(4)(2).
Freelancers and science communicators: you are likely a deployer
The AI Act distinguishes between providers (those who develop AI systems) and deployers (those who use AI systems to produce content). If you use an AI tool to generate or manipulate content that you then publish, you are a deployer. Article 50(4) applies to you.
The “purely personal, non-professional” exemption in Article 2(10) of the AI Act does not apply if you earn income from your content on a regular basis. It also does not apply if you do freelance, occupational, trade, or business activity. The Guidelines state this directly: any activity through which a person gains economic benefit on a regular basis is a professional activity. The Guidelines name freelancers explicitly.
Science communicators who work under institutional affiliation also fall within scope, even part-time or on contract. Their work is professional and typically informs the public on matters of public interest. The Guidelines give AI-generated Christmas cards sent to relatives as an example of personal use. The personal use exemption does not apply to published professional content.
What a compliant AI disclosure must include under Article 50(5)
Article 50(5) sets the horizontal requirements that apply to every disclosure under Article 50(1) to (4). You must provide the information in a clear and distinguishable manner, and at the latest at the time of the first interaction or exposure.
“Clear” means noticeable and easy to understand, including for persons with accessibility needs. “Distinguishable” means easy to identify as separate from other information and the surrounding environment. You must meet both conditions at the same time. A disclosure that is technically present but visually merged with body text, or buried below the fold, does not meet the standard.
For text content, “first exposure” means the start of the publication, at the moment a reader first sees it. The Guidelines state directly that a disclosure placed at the end of a piece does not comply with Article 50(5). The same rule applies to video: place the disclosure at the start, not in the end credits.
What does not count as an AI disclosure under the EU AI Act
The Guidelines are specific about what fails the standard. None of the following items, on their own, satisfy the obligation:
- Disclosures contained only in your terms and conditions, a linked URL, or supporting documentation. These can complement an in-content disclosure, but cannot replace it.
- Machine-readable markings, such as metadata or watermarks, that users cannot perceive without technical tools.
- Unclear or ambiguous signals, including generic references to “assistant” or vague language about using technology.
- Technical descriptions alone. A statement that your content “uses LLMs” or was “created with generative AI” does not disclose the artificial origin of the specific content in front of the reader.
- Information hidden under layers of menu options on your website or platform.
For live broadcasts, a one-time disclosure at the start may not be enough if viewers can join midway. In these cases, use periodic or persistent disclosure throughout the broadcast.
Where to place your AI disclosure: blog posts, social media, video and podcasts
Blog posts: Place the disclosure visibly at the top of the article, before the first paragraph of content. This is the point of first exposure for your reader. A disclosure at the end of a long post does not meet the requirement, no matter how detailed it is.
Social media posts: Place the disclosure at the start of the caption or post text, before any substantive content. The Guidelines confirm that AI-generated posts on matters of public interest are in scope, for example public safety warnings on a social media profile. The disclosure must appear before the reader reaches the content, not after it.
Video content: Display a visible text label or overlay at the start of the video, before your substantive content begins. The Guidelines state directly that disclosure “as part of end credits does not comply with Article 50(5) AI Act.” For live-streamed content, add periodic on-screen reminders. A single disclosure at the start is likely not enough for viewers who join after the broadcast starts.
Podcast show notes: Place the disclosure at the top of the show notes or episode description page, before the episode summary. If AI-generated content appears in the spoken episode, add a spoken disclosure at the start of the recording. The same logic of first exposure applies to audio.
The editorial responsibility exception: do you qualify?
Article 50(4) subparagraph 2 includes an exception. If a person or organization gives the AI-generated text human review or editorial control, and a person or legal entity holds editorial responsibility for the publication, the disclosure obligation does not apply to that text.
You must meet both conditions. “Human review” means deliberate examination of the substance of the content by someone with relevant competence and professional judgment. It is not a spell-check, a cursory read-through, or review by another AI system. “Editorial responsibility” means a specific person or entity holds the ultimate legal responsibility for what is published. You must make their identity and contact details publicly available, for example on a website’s terms and conditions page or in a publication’s colophon.
The Guidelines are direct about what does not qualify: superficial or procedural checks, the mere existence of an editorial policy, and cursory approval without substantive engagement. If you write, lightly review, and publish your own AI-assisted content as a solo creator, the exception is unlikely to apply to you.
Copy-paste AI disclosure templates for every content type
The following templates are based on the legal language in Article 50(4) of the AI Act and the Commission’s draft Guidelines. They are not officially prescribed wording. The Guidelines do not mandate specific phrasing. They reflect the core requirement: state clearly, at the point of first exposure, that the content was artificially generated or manipulated.
Blog post — fully AI-generated:
This article contains content that has been artificially generated or manipulated using AI tools.
Blog post — AI-assisted (partial):
Parts of this article were artificially generated or manipulated using AI tools.
Social media post (start of caption):
[AI-generated] This post contains text that has been artificially generated using AI.
Video (on-screen text overlay at the start):
This video contains content that has been artificially generated or manipulated using AI.
Podcast — spoken disclosure (start of episode):
This episode contains content that was artificially generated or manipulated using AI tools.
Podcast show notes (top of description):
Note: These show notes contain content that has been artificially generated or manipulated using AI.
Adapt the wording to your context. Keep the disclosure clear, distinguishable, and placed before your substantive content. Do not move it to end credits, footers, or linked policy pages.
The EU’s official AI-content icons
Alongside the text-based disclosure obligations above, the European Commission published a set of standard icons for labelling AI-generated content. The Commission developed these icons to support compliance with Article 50, as part of the Code of Practice on marking and labelling AI-generated content. Details: digital-strategy.ec.europa.eu — EU icons for labelling AI-generated content.
There are three variants:
- Basic icon. Signals that AI took part in creating the content. The surrounding label can vary (for example, custom wording or an interactive layer).
- Fully AI-generated. For content created entirely by AI, with no human-made elements beyond the prompt.
- Partially AI-modified. For pre-existing human content that AI altered.
Each icon comes in black, white, and 50%-transparency versions, in SVG and PNG. You can use the icons without attribution to the Commission.
Important nuance: the icon itself is optional. What is mandatory is the underlying disclosure obligation under Article 50(5) (clear, distinguishable, present at first exposure), described above. The icon is a recognized way to meet that standard visually, not a separate legal requirement. Code of Practice signatories commit to specific placement rules for the icon. Everyone else can adopt it voluntarily, as a faster and more visually distinguishable alternative or complement to a text disclosure line.
The same placement rule applies: the icon must be clearly perceivable at the latest at the time of first exposure, and must stay visible if the content is reshared or downloaded.
Placement mechanics: embed the icon directly within the content itself. Do not just link to it or reference it elsewhere on the page, except for evidently creative works. No overlapping element (a cookie banner, a popup, a sticky header) can obstruct it.
Accessibility requirements: the Commission’s guidance also covers these points. Text next to the icon must use plain language and avoid jargon or abbreviations beyond “AI” itself. The icon needs alt text or an ARIA label for screen readers. If you use the interactive second-layer version (for example, a tooltip or expandable label), users must be able to navigate it by keyboard and assistive technology. Where content is likely to reach users with cognitive difficulties, extend visibility beyond a brief flash.
How the icons relate to the Code of Practice: the icons formally implement Section 2 of the Code of Practice on Transparency of AI-Generated Content. Code of Practice signatories commit to following its placement specifications. If you are not a signatory, you can still use the icons freely, but doing so does not by itself signal Code adherence. The same source page also notes narrower disclosure exceptions beyond editorial review: evidently artistic, creative, satirical, or fictional works, and content whose disclosure is restricted by law (for example, for criminal investigation or prosecution purposes).
What you can do before August 2026
The transparency obligations under Article 50 apply from 2 August 2026. They apply to all in-scope AI systems and content in use on that date, not only to content created after it.
Audit your existing content now. This gives you time to build disclosure into your standard publishing workflow before the deadline. It also gives you the chance to assess whether any of your content can qualify for the editorial responsibility exception, and to put genuine substantive review processes in place if it does.
Take three actions now. Identify which of your regular content types fall within the scope of Article 50(4). Add a disclosure statement to your blog post, video, and social media publishing templates. Assess your editorial process honestly against the standard the Guidelines set. A light read-through before you press publish is not enough.
Related reading on The Science Talk
This guide accompanies the EU AI Act Article 4 explainer on The Science Talk, which covers the AI literacy obligations that research institutes must meet. It gives useful context for science communicators who work in institutional settings.
Want more guides like this? Browse all AI guides or get in touch →
Found this useful? Share it or explore more guides.