PageWeave / Docs

Platform

Markdown

AI Content Disclosure

Disclaimer: This page is general information about regulations — not legal advice and not an assessment of your specific situation. Rules vary by jurisdiction and change over time. Do not act on this page alone: for any compliance decision, consult a qualified lawyer. Full disclaimer at the bottom.

When a website must disclose AI-generated content under the EU AI Act — and when it doesn't. Label rules, snippets, and primary sources.

TL;DR: No EU law requires a "built with AI" badge just because an AI agent built your site. Disclosure duties attach to specific content: AI chatbots, deepfake media, and AI-generated text published to inform the public on matters of public interest. Other regimes (California, China) have their own rules — check what applies to your audience.

What applies, and since when

  • Article 50 of Regulation (EU) 2024/1689 (AI Act) applies since 2 August 2026 (Article 113; Article 50 explorer view).
  • The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force 27 July 2026) delayed most high-risk obligations — Annex III systems to 2 December 2027, Annex I products to 2 August 2028 — but did not change Article 50. Transparency applies now.
  • Providers of generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to bring machine-readable marking (Article 50(2)) into line. This binds AI system providers, not website owners.
  • No retroactive labelling: per the Commission Guidelines, media counts from its generation date; public-interest text counts from its publication date — text generated before 2 August 2026 but published after must be labelled.
  • Penalties for transparency breaches: up to €15 million or 3% of worldwide annual turnover, whichever is higher (Article 99(4)).

Who owes what

Duty Who What
Chatbot notice — Art. 50(1) Provider AI systems interacting directly with people must say so, unless obvious to a reasonably well-informed, observant and circumspect person
Machine-readable marking — Art. 50(2) Provider of the generative AI system Synthetic outputs marked and detectable (metadata, watermarks)
Emotion recognition / biometric categorisation — Art. 50(3) Deployer Inform exposed persons
Deepfake & public-interest text labels — Art. 50(4) Deployer Visible disclosure
Form of all disclosures — Art. 50(5) Both Clear, distinguishable, at first exposure, accessible

For a website, you are almost always the deployer (the site owner). The model provider (OpenAI, Anthropic, Google, …) carries the marking duties. If you embed a third-party chatbot on your site, make sure the AI notice actually appears on your surface.

The three triggers that matter for websites

  1. AI chatbot or assistant on the site. Visitors must be told they are interacting with AI at the first interaction. The "obvious from the context" exception is read narrowly by the Commission — a customer-support bot does not qualify.
  2. Deepfake media. AI-generated or manipulated image, audio, or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic. Text is not a deepfake under the definition (Article 3(60)).
  3. AI-generated text published to inform the public on matters of public interest. News-like text must be labelled as AI-generated — unless the editorial-responsibility exception applies (next section).

Not triggers: a site built by an AI agent, its HTML/layout, ordinary marketing copy, product descriptions, or AI-assisted editing that doesn't substantially alter content. A blanket "this site was made with AI" badge is not required — don't add one unless you want to.

The editorial-responsibility exception

Article 50(4) spares public-interest text that "has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication." This is not automatic — the Code of Practice expects deployers relying on it to document: who holds editorial responsibility (name, role, contact details) and the organisational resources behind the review. If you rely on the exception, be able to evidence it.

How to label

From the Code of Practice and Guidelines:

  • Perceivable without clicks, tools, or dedicated actions — a visitor must see or hear the disclosure passively.
  • At the latest at first exposure — before or as the content is consumed.
  • Text: label above/at the top of the text, near the headline, or in the colophon.
  • Media: label on or next to the image/video; for video also at the beginning and at regular intervals; no intervening overlays.
  • Never only in the footer, terms of service, end credits, or behind menu layers.
  • Accessible: conform to applicable accessibility requirements — WCAG 2.1 AA is the reference standard (with EN 301 549 for in-scope services).
  • Artistic, creative, satirical, or fictional works: disclosure is limited to a manner that does not hamper display or enjoyment (adjacent note, credits).
  • EU icons: optional, free to use without attribution, and often the clearest signal — see Official EU icons below. Using them is optional; labelling itself is not.

Snippets

For PageWeave sites (Tailwind + daisyUI included, no JS needed):

Public-interest text label — place above the article:

<p role="note" class="badge badge-warning badge-outline">
  AI-generated — created with AI and published to inform on matters of public interest
</p>

Chatbot first-interaction notice — visible when the chat opens:

<div role="status" class="alert alert-info text-sm">
  You are chatting with an AI assistant, not a human. It can make mistakes.
</div>

Media label:

<figure>
  <img src="/assets/…" alt="Describe the image">
  <figcaption class="mt-2 text-sm opacity-80">
    <span class="badge badge-warning badge-outline">AI-generated</span>
    This image was artificially generated; it does not depict a real person or event.
  </figcaption>
</figure>

Official EU icons

The Commission publishes three icons: AI GENERATED (wholly AI-generated content), AI MODIFIED (real content altered by AI), and a generic AI (AI involvement with a custom text label). Picking the wrong one is a factual misstatement — match the icon to how the content was made. User testing by the Commission found pairing the icon with a short text label works best.

AI GENERATED AI MODIFIED AI

Hosted copies for agents and quick download (byte-identical to the official release, Code of Practice Annex I, 10 June 2026):

Icon Black White (transparent)
AI GENERATED SVG SVG
AI MODIFIED SVG SVG
AI SVG SVG

All variants (black, white, 50% transparency; PNG) ship in the official SVG ZIP / PNG ZIP; display specifications live on the Commission icons page. Key rules: clearly perceivable at first exposure, placed where no overlay covers it, directly embedded in the content so it survives resharing or downloading.

The icons are free to use without attribution. Using them does not signal adherence to the Code of Practice and does not establish legal compliance by itself — the Article 50 labelling duty applies regardless.

For a website, upload the icon as a site asset once (via the dashboard or upload_asset url:), then reference the asset URL:

<figure>
  <img src="/assets/…" alt="Describe the image">
  <figcaption class="mt-2 flex items-center gap-2 text-sm">
    <img src="/assets/<your-site-asset>/eu-ai-generated-black.svg" alt="AI GENERATED" class="h-6">
    <span>AI-generated image — does not depict a real person or event.</span>
  </figcaption>
</figure>

Other regimes

This page focuses on the EU. Depending on audience and jurisdiction, other rules may apply — check them for your site:

  • California SB 243 (companion chatbots, in force 1 January 2026): if a reasonable person could be misled into thinking they're talking to a human, a clear and conspicuous "this is AI" notification is required — plus extra duties for known minors and a private right of action (greater of actual damages or $1,000 per violation). Ordinary customer-service bots are excluded. (bill text)
  • California SB 942, the AI Transparency Act (operative 2 August 2026): duties fall on "covered providers" of generative AI systems with over 1,000,000 monthly users — and from 1 January 2027 on large online platforms (2,000,000+ users) — not on ordinary websites. (statute)
  • China — Measures for the Identification of AI-Generated Synthetic Content (effective 1 September 2025, with mandatory standard GB 45438-2025): the strictest regime — visible labels on AI text (start/end/middle), prominent labels on images and video start frames, plus embedded metadata; platforms must verify and propagate labels. Relevant when serving users in China. (official text, MIIT announcement)

Sources

Disclaimer

This page is general information, not legal advice.

  • General information only. This page describes legal rules in general terms and cites primary sources. It does not examine, assess, or apply the law to your specific website, content, business, or situation — any such individual assessment would be legal advice, which we do not provide.
  • No professional relationship. Reading this page, or using PageWeave, does not create an attorney-client, advisory, or any similar professional relationship with PageWeave. We are not a law firm and are not admitted to practise law.
  • No reliance. Do not act, or refrain from acting, based on this page alone. Whether any labelling or disclosure duty applies to you depends on facts — your content, your audience, your jurisdiction — that only a professional review can establish.
  • No warranty of currency or completeness. Laws, guidelines, and their interpretation by courts and authorities change over time. We make no guarantee that this page is accurate, complete, or up to date at the time you read it.
  • Consult a qualified lawyer. For any concrete compliance decision — including whether, how, and where to label content — seek advice from a lawyer qualified in your jurisdiction. Nothing on this page replaces that advice.
  • Limitation of liability. To the extent permitted by law, PageWeave accepts no liability for decisions made, or actions taken, in reliance on this page.