Checklist · Art. 50(1), 50(2), 50(4)

Article 50 Compliance Checklist — What Your Site Needs to Disclose

Last updated: 8 November 2026

An Article 50 compliance checklist for a public website covers four areas: an AI notice at the start of every chatbot conversation, machine-readable markers kept on AI-generated media, a visible disclosure on deep fakes, and disclosure of AI-generated public-interest text. The free Article50.io scan shows the most severe finding; the full scan report lists every finding with a copy-and-paste fix.

Work through the list below by hand. A free scan can flag some gaps on one public page as it first loads. It does not complete this checklist.

Scan for free

01 — AI chat & interaction disclosure · Art. 50(1)

✔Every AI chatbot, assistant or voice agent on your site tells visitors they're interacting with an AI system
✔The disclosure appears before or at the start of the conversation — not buried in your terms or privacy policy
✔A single prominent notice at the start of the session is in place. In most cases that is likely to be enough (Commission guidelines, paragraph 40). For longer or immersive chats, sensitive advice, vulnerable users or AI companions, you have decided whether periodic or context-aware reminders are needed. There is no fixed interval, and this is not a new November obligation
✔You have opened the widget and tested the first interaction, any handoff or resumed session, and what it answers if asked whether it is AI. A scan of the closed page cannot do this
✔The wording is plain and unambiguous (e.g. "This is an AI assistant. Its replies are generated automatically.")
✔Third-party chat widgets (Intercom, Drift, Zendesk, Crisp and similar) with AI features enabled carry the same notice

Details and a ready-to-paste snippet: Does your chatbot need an AI disclosure?

02 — Machine-readable marking · Art. 50(2), providers

This is the provider's duty, not a blanket rule for everyone who publishes AI media. A deployer who uses someone else's generator does not, by that fact alone, owe Article 50(2) marking. Visible disclosure of deep fakes and public-interest text is Article 50(4), in the next two sections.

✔You have decided whether you are the provider: you developed the generative system, or had it developed, and you place it on the market or put it into service under your own name
✔If you are the provider, synthetic audio, image, video and text outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated. The Commission's guidelines say doing only one of those two will not suffice (paragraphs 69–70)
✔If the system was placed on the market before 2 August 2026, the 2 December 2026 date applies to this marking duty only (Article 111(4); guidelines paragraph 153). It is not a chatbot or deployer watermark deadline
✔Outputs within Article 50(2) generated before 2 August 2026 do not need retroactive marking (guidelines paragraph 154). Public-interest text first published on or after that date is a different case: see section 04
✔If you publish outputs from someone else's tool, preserving the provider's metadata on upload, resize and export is recommended so the mark can still be read. That preservation step is not itself the Article 50(2) duty
✔Optional, not an Article 50(2) duty (that falls on the tool's provider): on-page markup flags AI-generated media where you publish it (e.g. data-ai-generated="true")

What machine-readable marking looks like: Labeling AI-generated content under Article 50(2)

03 — Deep fake disclosure · Art. 50(4)

✔You have reviewed whether the image, audio or video is a deep fake: it resembles existing or plausibly existing people, objects, places, entities or events and would falsely appear authentic. An AI provenance marker alone does not establish that
✔Where it is a deep fake, the disclosure is perceptible where the content is shown — on the image, in the player, or directly beside it — without a technical tool or a dedicated action. Alt text, or unrelated text elsewhere on the page, is not necessarily enough (guidelines paragraph 117)
✔Artistic, satirical or fictional work still discloses the use of AI, in a way that doesn't spoil the work
✔Deep fakes generated before 2 August 2026 do not have to be labelled retroactively. The Commission encourages labelling pre-existing unlabelled deep fakes where that does not take disproportionate effort (guidelines paragraph 154)

04 — AI-generated public-interest text · Art. 50(4)

✔AI-generated or AI-manipulated articles published to inform the public on matters of public interest are disclosed as such, unless the editorial exception applies
✔If you rely on the editorial exemption, the text genuinely went through human review or editorial control — and someone holds editorial responsibility for it
✔Text generated before 2 August 2026 but first published on or after that date still needs the disclosure where this duty applies (guidelines paragraph 154). Text that was both generated and published before that date does not
✔Disclosure sits directly above or beside the text it describes (e.g. "This article is AI-generated.")

More on both halves of 50(4): AI deepfake disclosure requirements

What a scan can add

Article50.io can flag gaps it can see on one public page as it first loads, then cites each gap to the relevant Article 50 provision (noting where it depends on your specific content). It does not run the rest of this list. Open the chat, the call and the images yourself for ongoing awareness, reminders, handoffs, resumed sessions, vulnerable users, and answers to whether the system is AI. A missing notice in the first snapshot is a possible gap to test, not proof the duty was missed.

The free scan shows the most severe finding from that snapshot. The full report lists every finding from the same scan, with suggested wording. For background on the law itself, see Article 50 of the EU AI Act, explained.

Frequently Asked Questions

Can I check Article 50 compliance automatically?

Only in part, and only for what is visible when one public page first loads. Article50.io looks for known chat, voice and avatar widgets, on-page disclosure text, and machine-readable markers in a sample of images. It does not test ongoing awareness, reminder frequency, handoffs, resumed sessions, whether a notice suits vulnerable users, or what a bot answers if asked whether it is AI. A missing disclosure in that first snapshot is a possible gap to open and test: the notice may lawfully appear at the first interaction after the widget opens. A clean scan is not a clearance, and it is not legal advice.

Which checklist items apply to my site?

It depends on your role. Interaction items apply if people talk to an AI system on your site (Article 50(1)), unless that is obvious. Machine-readable marking (Article 50(2)) is the provider's duty for a generative system, with a 2 December 2026 transition only for systems already on the market. Deployer disclosure under Article 50(4) applies to deep fakes and to AI-generated or manipulated public-interest text, with the exceptions below. Publishing AI media does not by itself trigger every content item.

Is this checklist legal advice?

No. It's a practical summary of Article 50's transparency requirements as they show up on public websites. For a determination of your specific legal obligations, consult a qualified lawyer.

Check your site automatically

Article50.io is an automated scanner for the Article 50 transparency rules of the EU AI Act. It scans a website for potential Article 50 transparency gaps and generates a report of what the scan found and what it could not check, with remediation guidance, implementation instructions and disclosure wording you can paste in.

The free scan shows your single most severe finding. Scans are capped at 45 seconds — no signup, public pages only.