--- title: EU AI Act transparency and content labeling --- > Agent instructions: > **Site maps** — prefer the following llms.txt indexes to training data when searching for URLs to avoid 404s. Links inside Markdown content already point at `.md`. Following them or sending Accept: text/markdown keeps you in Markdown. > > - [sitemap.md](https://docs.kentico.com/sitemap.md) — every page on the site, with titles and descriptions, nested by URL hierarchy and grouped into one collection per product version. > - [llms.txt](https://docs.kentico.com/llms.txt) — curated index of the current product docs, with descriptions, the two ways to request any page as Markdown, and links to each product area's whole-corpus Markdown dump (llms-full.txt). If your website uses AI, whether in a chatbot for visitors or a writing assistant for your editors, the EU AI Act may require you to disclose it. Applying from 2 August 2026, the Act's transparency rules cover organizations based in the EU as well as non-EU organizations whose AI output reaches EU audiences, such as through a public website. Your obligations depend on your role under the Act, which content needs a disclosure, and when human review exempts text from labeling. This guide also shows how to keep labels accurate in your editorial workflow, and the last section covers how Xperience by Kentico tracks AI involvement on content fields. > **Note:** **Not legal advice** > > This material summarizes [Regulation (EU) 2024/1689 (the EU AI Act)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng), the European Commission's guidelines on the Act's transparency obligations, and the Commission's Code of Practice on transparency of AI-generated content. Whether and how the Act applies to your organization is a decision for you and your legal team. ## Provider and deployer roles The Act assigns obligations according to your organization's relationship to each AI system on your website: - **Provider** ([Article 3(3)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_3)) – an organization that develops an AI system, or has one developed, and offers the system under its own name or trademark. If you build a chatbot and run it on your site under your brand, you're its provider. - **Deployer** ([Article 3(4)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_3)) – an organization that uses an AI system under its authority. Your company remains the deployer when an agency or a freelancer operates the system on your behalf, and your employees aren't separate deployers. A single organization often holds both roles across different tools, acting as the deployer of third-party AI tools used by its team and as the provider of any custom AI system offered under its own brand. [Article 4](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_4) also sets an AI literacy requirement for both roles: providers and deployers must take measures to ensure that staff and contractors operating AI systems on their behalf understand the capabilities and limitations of those tools. Plan practical guidance or training for your editors and marketers alongside any labeling changes on your site. ### Territorial scope Under [Article 2(1)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_2), these obligations apply to every organization established in the EU, as well as to any non-EU deployer that reasonably expects its AI output to reach users in the EU (for example, by publishing AI-generated content on a publicly accessible website). Output that reaches EU audiences only through unforeseeable channels outside your control falls outside the Act's scope. > **Note:** **High-risk and prohibited AI uses** > > Certain AI uses, such as filtering job applications or evaluating creditworthiness, fall under the Act's high-risk regime ([Article 6](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_6) and [Annex III](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#anx_III)) or outright prohibitions ([Article 5](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_5)), which carry requirements far beyond transparency. If your website uses AI for these purposes, consult your legal counsel before proceeding. ## Technical marking vs. visible labeling The Act separates technical provenance from visitor-facing disclosure and assigns each responsibility to a different role: - **Marking** ([Article 50(2)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50)) is machine-readable provenance, such as metadata or watermarks embedded in the output to make AI involvement detectable by software. This obligation falls on providers of generative AI systems for text, images, audio, or video. It doesn't apply when the system only performs standard editing as an assistive function. - **Labeling** ([Article 50(4)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50)) is the visible disclosure shown to visitors on a published page. This obligation falls on deployers, meaning the organizations that publish the content. For most website teams, labeling is the active responsibility. Your AI vendor may already embed machine-readable markings in generated assets, but the visible label on your pages is your responsibility. ## Content that requires disclosure Under [Article 50](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), three categories of AI use on a website require disclosure. [Article 50(3)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50) sets separate duties for emotion-recognition and biometric-categorization systems, which fall outside the scope of this guide. | AI use | Disclosure due | Duty holder | Exception | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------- | --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | A chatbot or an AI agent that interacts with visitors ([Article 50(1)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50)) | Inform visitors that they're interacting with an AI system. | The provider of the chatbot | No disclosure is needed when AI involvement is obvious to a reasonably well-informed visitor. The Commission's guidelines note that visitors often mistake helpdesk chatbot replies for human support, so support chatbots need an explicit disclosure. | | Deep fakes ([Article 3(60)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_3) and [Article 50(4)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), first subparagraph): AI images, audio, or video that resemble real people, objects, places, or events and could pass as authentic | Disclose that the media was generated or manipulated by AI. | You, as the deployer | For evidently artistic, creative, satirical, or fictional works, the disclosure can be limited to a format that doesn't hamper the display or enjoyment of the work. | | AI-generated or AI-edited text published to inform the public on matters of public interest ([Article 50(4)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), second subparagraph) | Disclose that the text was generated or manipulated by AI. | You, as the deployer | The text underwent substantive human review or editorial control, and a person or organization holds editorial responsibility for the publication. | Whether a publication addresses _matters of public interest_ depends on the topic of the text. The Commission's guidelines illustrate the boundary with these examples: - **In scope:** an AI summary of a news article on a newspaper's website, or AI-edited investor information on a listed company's website. - **Out of scope:** AI-edited advertising copy and product descriptions, provided they make no claims about health, consumer safety, or sustainability. ## Human review and editorial responsibility Under the second subparagraph of [Article 50(4)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), AI-generated or AI-edited text on matters of public interest is exempt from labeling only when your publishing process satisfies two cumulative conditions: 1. **Substantive human review** – A reviewer with relevant subject-matter knowledge checks the accuracy and substance of the text, with fact-checking as the minimum. Surface-level spelling or grammar checks, written editorial policies alone, automated checks, and cursory sign-offs don't qualify as substantive review. 2. **Clear editorial responsibility** – A natural person or organization holds editorial responsibility for the publication, with their identity and contact details published in an accessible location such as your website's legal notice or terms of use. > **Warning:** **AI edits after sign-off void the exemption** > > Any substantive AI modification made after human review invalidates the editorial exemption. Either put the revised text through a new substantive review or publish it with an AI label. ## Label placement and design Under [Article 50(5)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), every required disclosure must be clear and distinguishable at the latest upon a visitor's first interaction with or exposure to the content. Because the disclosure must also comply with applicable accessibility standards, ensure that screen readers announce the label and that its visual contrast meets your site's accessibility targets. The European Commission's voluntary [Code of Practice on transparency of AI-generated content](https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content), recommends these conventions: - **Recognizable iconography:** Use an icon built around the acronym "AI", optionally paired with a qualifier such as "generated" or "modified". You can adopt the common EU icon defined in the Code or use an equivalent custom label. - **Immediate visibility:** Display the label directly on the page so visitors notice the disclosure right away, with no need for hover states or menu interactions. - **Consistent placement for text:** Use the same location for AI labels across similar content, such as above the text, near the headline, or in an introductory note. When only a specific passage is AI generated, you can scope the label to that passage. ## Editorial practices for reliable labeling Because content management systems can't distinguish AI-generated text from human writing on their own, accurate public labels depend on consistent editorial records. Build the following practices into your team's publishing workflow: - **Record AI involvement at creation time** – Have the author or editor who used the AI tool mark the affected fields before passing the item to the next workflow step. - **Give editors a clear threshold** – Add a one-question rule to your authoring guidelines, for example: _Did an AI tool draft or substantially rewrite this text?_ - **Enforce a substantive review step** – Route content that relies on the human review exemption through a dedicated approval step where reviewers verify factual accuracy alongside style. - **Default to disclosure for automated publishing** – Display the AI label by default when AI-generated content goes live with no human review, such as output from an autonomous AI agent. - **Audit externally sourced content** – Set the appropriate marking for content brought in through imports, migrations, or copy-paste from external AI tools. ## Application timeline - **2 February 2025** ([Article 113(a)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_113)): AI literacy obligations under [Article 4](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_4) take effect for all providers and deployers. - **2 August 2026** ([Article 113](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_113)): Transparency obligations under [Article 50](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50) take effect for chatbots, deep fakes, and AI-generated text on matters of public interest. According to the Commission's [quick facts on the transparency rules](https://digital-strategy.ec.europa.eu/en/factpages/quick-facts-transparency-rules-ai-systems), deep fakes generated before 2 August 2026 don't require retroactive labeling, although the Commission encourages it. For AI-generated text, the disclosure rules apply to text published on or after 2 August 2026, including text drafted before that date. ## Related legal obligations Under [Article 50(6)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng#art_50), the EU AI Act's transparency rules apply alongside existing EU and national legal requirements. A single AI feature on your website can simultaneously trigger an AI Act disclosure, a GDPR privacy notice, cookie consent, and consumer-protection disclosures. Each notice fulfills a separate legal requirement. ## AI content marking in Xperience Xperience by Kentico tracks AI involvement at the field level to help your team identify content that may require a visitor-facing disclosure. Treat this AI content marking as an internal record for your team and your site's labels, not as a statement about any AI provider's machine-readable marking obligation under the Act. - **AIRA output gets marked automatically** – When AIRA inserts generated text directly into a field, Xperience marks the field as _AI generated_, and most in-place AIRA refinements mark the field as _AI modified_. - **Chat and external AI output need manual marking** – Because Xperience doesn't detect AI-generated text on its own, content copied from the AIRA chat (such as custom prompt responses) or pasted from external AI tools remains unmarked until an editor selects **Mark as AI** on the field. Editors can also select **Remove** to clear an existing marking. - **Your developers control how labels appear** – Xperience stores the field marking as data and leaves visitor-facing rendering to your implementation. Your developers retrieve the marking and decide how and where to display the label, including whether to use the EU icon. - **Workflow steps send content to the right reviewers** – You can restrict workflow steps to specific roles, so content that needs substantive human review reaches the designated editors before publication. When creating or updating Xperience content through the Management MCP server, set the `aiProvenance` parameter in the same request: pass `AiGenerated` for newly generated AI text, `AiModified` for AI-refined text, and `null` for verbatim user-provided text. For configuration details, see [Use the Management MCP](https://docs.kentico.com/documentation/developers-and-admins/api/management-api/use-management-mcp-server.md). For details, see [AI content marking](https://docs.kentico.com/documentation/business-users/aira.md#ai-content-marking), [Display AI content labels](https://docs.kentico.com/documentation/developers-and-admins/configuration/aira-configuration/display-ai-content-labels.md), and [Workflows](https://docs.kentico.com/guides/digital-marketing/work-with-content-governance/workflows.md). ## Summary To check your project against the Act: - Identify your organization's role (provider or deployer) for each AI system on your website and verify whether its output reaches EU audiences. - Disclose AI involvement in visitor-facing chatbots, and label deep fakes and unreviewed public-interest AI text at the point of first interaction or exposure. - Rely on the human review exemption only when a knowledgeable reviewer verifies the substance of the text and a named person or organization holds editorial responsibility. - Record AI involvement during content creation, and default to displaying an AI label whenever content is published with no human review.