When Must You Disclose ChatGPT Content Under the EU AI Act? A Scenario-by-Scenario Breakdown

ChatGPT has become integral to content creation in European businesses, powering everything from blog posts to social media campaigns. However, the EU AI Act introduces specific transparency obligations that many content creators haven’t fully considered.

Article 50(4) of the EU AI Act requires deployers to disclose AI-generated text only where it is “published with the purpose of informing the public on matters of public interest,” and it exempts text that has undergone human review or editorial control where a person or organisation holds editorial responsibility. That is far narrower than most marketing and communications use of ChatGPT. The Code of Practice on Transparency of AI-Generated Content offers further guidance on implementation.

The challenge for content teams is not only understanding the legal requirement but also pinpointing when disclosure applies. Some scenarios clearly demand disclosure, while others fall into grey areas requiring careful analysis. This article offers a scenario-by-scenario breakdown to guide your compliance decisions.

Article 50 of the EU AI Act sets transparency requirements for AI-generated content. It requires deployers to disclose AI-generated text that is “published with the purpose of informing the public on matters of public interest.” Text with a purely commercial or entertainment purpose is not covered.

These Article 50 requirements extend beyond high-risk systems to general-purpose models like ChatGPT used in public-facing communications. The trigger is the content’s purpose — informing the public on a matter of public interest — not merely whether the text could be mistaken for human-written.

Key triggers include:

The obligation applies regardless of whether the AI system is classified as high-risk elsewhere in the Act. But where any of those conditions is missing — the content is an advert, a product description, or an edited post someone is editorially responsible for — the Article 50 text-disclosure duty does not apply.

ChatGPT is classified as a general-purpose AI model under most EU AI Act classifications, not as a high-risk system. However, this doesn't exempt users from transparency obligations when they deploy it for public-facing content.

This distinction matters: although ChatGPT itself may not fall under the strictest AI Act regulations, businesses deploying it must comply with Article 50's transparency requirements. The law focuses on the deployment context, not the model's risk classification.

Foundation models like ChatGPT pose compliance challenges due to their versatility. The same model can generate internal brainstorming notes (no disclosure required) or published blog posts (disclosure required). Content teams must therefore assess each use case individually rather than applying blanket rules based on the model's general classification.

The upcoming EU AI Act deadlines emphasize the need to establish clear internal procedures for determining when ChatGPT-generated content requires disclosure across various deployment scenarios.

Being private takes content out of scope entirely — but being public is not, on its own, enough to require disclosure: publication is only the first of three cumulative conditions (published + informing the public + on a matter of public interest), and the editorial-responsibility exemption can still remove the duty. "Published to inform the public" refers to any material accessible to external audiences via websites, social media, marketing materials, or other public channels.

Published content that can fall in scope — but only where it also informs the public on a matter of public interest and no one holds editorial responsibility for it:

Private content generally exempt from disclosure includes:

The determining factor is the intended audience, not the medium. For example, an email newsletter sent to thousands of subscribers is public communication, whereas a personalized reply to a specific customer remains private correspondence.

Content teams should define clear protocols to identify when ChatGPT-generated content shifts from private drafts to public communications. This threshold triggers disclosure obligations.

Blog Posts and Website Content: In scope only where the post genuinely informs the public on a matter of public interest and no human holds editorial responsibility for it. A blog with a named author or editor who signs off the copy is exempt under Article 50(4); a purely automated public-interest article is not.

Marketing Copy and Advertisements: Generally out of scope for Article 50. The Commission’s guidelines cite “AI-manipulated text that is part of a company’s advertisement” as an example of text that is not covered (unless it makes health, consumer-safety or sustainability claims). FTC/ASA advertising rules and platform policies may still apply.

Social Media Posts: The Article 50 text duty bites only for posts that inform the public on a genuine public-interest matter with no editorial responsibility; posts about your own business or promotions sit outside it, though platform policies may still require an AI label.

Product Descriptions: Not covered by the Article 50 text obligation — product copy informs a purchasing decision, not “the public on a matter of public interest.” (Consumer-protection law still requires it to be accurate and non-misleading.)

Customer-Facing Emails: A marketing newsletter is commercial, not public-interest, text and is generally out of scope. The duty could apply only where an email genuinely informs recipients on a public-interest matter without editorial responsibility.

Internal Documents: Generally exempt from disclosure, as they’re not published for external audiences—e.g., internal reports, meeting notes, planning documents, and draft materials.

Press Releases: Potentially in scope where the release addresses a genuine public-interest matter — but a release issued by an identifiable organisation normally carries editorial responsibility, which engages the Article 50(4) exemption.

The EU AI Act includes several carve-outs. The most important for text is the editorial-responsibility exemption: Article 50(4) does not apply where the AI-generated text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication — which exempts most professionally-run blogs, newsrooms and company sites where a named person signs off the content. Understanding these exceptions helps content teams focus disclosure efforts where they’re legally required.

Artistic and creative works—fictional stories, poetry, and artistic descriptions—generally don’t require AI disclosure when clearly presented as creative content. However, this exception doesn’t cover commercial content masquerading as artistic expression.

Satirical content and clear parody are also exempt when the humorous or satirical intent is obvious to readers. Apply this exception carefully, since marketing content using humor still serves commercial purposes.

The comprehensive guide to disclosure exceptions provides detailed analysis of additional exemptions, including research contexts and academic use cases where disclosure requirements may be modified or waived entirely.

Legally privileged communications maintain confidentiality protections that can override disclosure requirements in specific professional contexts, though this exception rarely applies to typical marketing and content creation scenarios.

Effective disclosure requires clear, accessible language that informs readers about AI involvement without disrupting content flow. Simple statements like “This content was generated with AI assistance” or “Parts of this article were created using AI technology” provide sufficient transparency for most scenarios.

Placement strategies vary by content type. For example, blog posts can include statements at the beginning or end; social media posts can use hashtags like #AIGenerated or #CreatedWithAI; and email newsletters can incorporate disclosures in headers or footers.

Structured data markup offers technical compliance for web content, but visible disclosures remain the gold standard. Many organizations combine both methods to ensure comprehensive compliance.

Industry best practices stress consistency across platforms. Establishing standard disclosure language and placement guidelines helps content teams maintain compliance while preserving editorial efficiency.

Content teams can assess disclosure requirements with this step-by-step framework:

Step 1: Is the content published or available to external audiences?
If no, disclosure generally isn’t required.

Step 2: Does it inform the public on a matter of public interest (not merely commercial offerings or your own business activities)?
If no, consider whether exemptions apply.

Step 3: Did ChatGPT contribute substantially—text, ideas, or structure—to the final content?
Minor assistance may not trigger disclosure.

Step 4: Has the text undergone human review or editorial control, with a person or organisation holding editorial responsibility? If yes, it is exempt under Article 50(4). (Artistic, satirical and fictional works are also outside scope.)
Apply exemptions carefully to avoid over-broad interpretation.

Key questions for assessment:

Quick reference checklist:
Published + informing the public on a matter of public interest + no editorial responsibility = disclosure required.

Under the EU AI Act, ChatGPT content disclosure is determined by whether content is published to inform the public, not by the AI system’s sophistication. Content teams need clear decision frameworks to assess disclosure requirements across scenarios while avoiding unnecessary compliance burdens on internal or exempt content.

The complete compliance checklist offers guidance for implementing systematic disclosure procedures across your content operations. Establishing these procedures now positions organizations for confident compliance as AI Act enforcement begins.

This article is general information, not legal advice.