Guide · Art. 50

EU AI Act Compliance Checker: What an Article 50 Check Should Actually Cover

Last updated: 8 November 2026

Searching for an EU AI Act compliance checker? The Commission offers a free official one: a questionnaire that helps you work out which AI Act rules may apply. For Article 50 on a website, a check should also look at the live page: AI chat notices, deep fake and AI text disclosures, and notice clarity. Article50.io scans that page.

This page is general information about checking a website against Article 50 of the EU AI Act, not legal advice.

Scan for free

Start with the official EU AI Act Compliance Checker

If you searched for an “EU AI Act compliance checker”, start with the free one. The European Commission runs the EU AI Act Compliance Checker on its AI Act Service Desk. The Service Desk says it “helps users understand which AI Act rules may apply to their AI system, including possible obligations for providers, deployers and other operators”. It is currently in its beta version.

It is a questionnaire. You answer questions about your AI system and it points you to the rules that may apply. It does not look at your website. That makes it a good first step for the big question (“which parts of the AI Act concern us?”), but it can't tell you whether the chat widget on your homepage tells visitors it is AI.

For that, you need a check of the page itself. The rest of this guide covers what such a check should look at under Article 50 of Regulation (EU) 2024/1689, which applies from 2 August 2026. For the background, see Article 50 explained.

What an Article 50 check on a website should cover

Article 50 sets transparency duties for certain AI systems. On a public website, these are the points a check should look at.

✔AI chat, voice agents and avatars. Under Article 50(1), providers must design AI systems that interact directly with people so that people are informed they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person. The Commission's Article 50 guidelines (para. 11) add that a company that builds a chatbot in-house and puts it into service under its own name is a provider.
✔Timing. Article 50(5) says the information must be given “at the latest at the time of the first interaction or exposure”. A notice only in a privacy policy, or one that appears after the conversation is under way, comes too late.
✔Deep fakes. Under Article 50(4), deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake must disclose that the content has been artificially generated or manipulated. For evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing that such content exists, in a way that does not hamper the display or enjoyment of the work.
✔AI text on matters of public interest. Also under Article 50(4), AI-generated or manipulated text published to inform the public on matters of public interest must be disclosed, unless it has undergone human review or editorial control and someone holds editorial responsibility for it.
✔AI image markers versus visible disclosures. Machine-readable markers in image metadata (such as an IPTC digital source type or a C2PA manifest) and a visible disclosure are different things. A good check notes when an image is marked as AI-generated but nothing on the page tells visitors so.
✔Clarity and accessibility. Article 50(5) requires the information to be “clear and distinguishable” and to “conform to the applicable accessibility requirements”. So a check should look at how the notice is shown: is it visible, readable, high enough in contrast, exposed to screen readers, and in the page's language?

There is no blanket duty to put a visible label on every piece of AI-generated content. Visible disclosure comes from Article 50(4), for deep fakes and public-interest text.

What the website owner can't check from the page

Some Article 50 duties are not visible on your site at all. Article 50(2) requires providers of AI systems that generate synthetic audio, image, video or text to mark the outputs “in a machine-readable format and detectable as artificially generated or manipulated”. That is the duty of the provider of the generative AI tool, not the website that publishes the output. A missing marker on your page does not tell you whether the tool's provider has met its duty. More on this in EU AI Act watermarking.

Article 50(3), on emotion recognition and biometric categorisation systems, is also hard to see from a public page. If your setup raises questions like these, the Article 50 questions page is a good next stop, and a lawyer is the right person for a firm answer.

Why form-based checkers miss live-site issues

Form-based checkers, including the official one, know what you tell them. That is fine for working out which rules may apply. It falls short on what visitors really see.

✔Forgotten widgets. A chat widget added by a marketing team two years ago, or loaded by a plugin, won't appear in your answers if nobody remembers it.
✔Vendor defaults. Many chat tools ship with a friendly name and avatar and no AI notice. Your answer may be “we use a chatbot”; the question is what the widget actually says when the page loads.
✔What visitors really see. A disclosure may exist in your settings but be hidden, tiny, low in contrast or invisible to screen readers.

In our automated scan of 59 popular EU websites with a chat widget (21–22 October 2026), 58 showed no AI disclosure at first view. The details are in the study.

How Article50.io scans the real page

Article50.io loads one public page in a real browser (Chromium), the way a visitor would. Each scan is capped at 45 seconds and needs no signup. It looks for:

✔AI-capable chat widgets from known vendors, and custom chat markup, with no AI disclosure when the page loads.
✔AI voice agents and AI avatars from known vendors with no AI disclosure.
✔AI-generated content markers. For images, it reads up to 8 of the page's main images and checks their metadata for an IPTC digital source type or a C2PA (Content Credentials) manifest saying the image was AI-generated. It reports an image when no label, caption, alt text or other text on the page says so (English wording). A missing marker is never a finding.
✔Where a disclosure exists, how clearly it is shown as the page loads: visibility, text size, contrast, exposure to assistive technology and language.

What it can't see: it does not open the chat or start a conversation, it does not crawl the whole site (one page per scan), and it cannot check whether your AI tool's provider marks its outputs under Article 50(2). Each result lists what the scan couldn't check. More detail on how the scan works.

The free scan shows the single most severe finding. The €499 scan report, a one-off payment with no subscription, covers every finding from the same scan, each with a ready-to-paste fix, plus a unique report ID and SHA-256 fingerprint, a “Checked” badge and one free recheck within 30 days. See pricing, or run a free scan.

How the options compare

Each option answers a different question, and many teams use more than one. For a longer look, see Article 50 checkers compared.

Form-based checkers, the official EU AI Act Compliance Checker, a consultant review and Article50.io compared
	Form-based checkers ($0-$99)	Official EU AI Act Compliance Checker (free)	Consultant review (a few hundred to a few thousand)	Article50.io (€499)
How it works	You answer questions about your AI setup	You answer a questionnaire on the Commission's AI Act Service Desk	A person reviews your setup; scope varies	We scan your site live
Catches a chat widget you forgot about	Only if you tell it about it	No. It doesn't look at your website.	Usually, if it's in scope	Yes
Checks the disclosure text visitors really see	No	No. It doesn't look at your website.	Usually, if it's in scope	Yes
Checks for machine-readable content markers	No	No. It doesn't look at your website.	Usually, if it's in scope	Yes
Fixes	Generic disclosure text	Not applicable	Usually written recommendations	Ready-to-paste fixes for each finding
After purchase	Report only	Not applicable	Depends on the engagement	"Checked" badge and one free recheck within 30 days
Best for	A quick first look at your obligations	Working out which AI Act rules may apply to you	A human review; scope varies	Knowing what's really detectable on your site

Honest scope: Article 50 only

Article50.io covers Article 50 of the AI Act and nothing else. It does not cover prohibited practices under Article 5, high-risk systems under Article 6 and Chapter III, general-purpose AI models, or the GDPR. The Commission's guidelines (para. 25) note that Article 50 duties apply alongside these: a system in scope of Article 50 may also be high-risk.

A scan reports what is detectable on one public page at one moment. A clean result is not a compliance clearance, and the scan report is not legal advice. Under Article 99(4), breaches of the Article 50 transparency duties can lead to fines of up to €15 million or, for an undertaking, up to 3% of total worldwide annual turnover, whichever is higher; for SMEs and small mid-caps, the lower of the two applies (Article 99(6) and (6a)). See EU AI Act fines for Article 50. For a firm view on your obligations, use the official questionnaire as a starting point and speak to a lawyer.

Frequently Asked Questions

Is there an official EU AI Act compliance checker?

Yes. The European Commission offers a free EU AI Act Compliance Checker on its AI Act Service Desk. It is a questionnaire, currently in beta, that helps you understand which AI Act rules may apply to your AI system. It does not look at your website.

Is Article50.io an EU AI Act compliance checker?

No. Article50.io is a scanner for Article 50 of the AI Act. It loads one public page in a real browser and reports what it detects, such as a chat widget with no AI disclosure. It does not assess the rest of the AI Act. A clean result is not a compliance clearance, and it is not legal advice.

What should an Article 50 website check look at?

Whether AI chat, voice agents and avatars tell people they are AI at the latest at the first interaction, whether deep fakes and public-interest AI text are disclosed, whether AI-marked images have a visible disclosure where one is needed, and whether any notice is clear, distinguishable and accessible, as Article 50(5) requires.

Can a website scan check Article 50(2) marking?

Not fully. Article 50(2) requires the provider of a generative AI system to mark its outputs in a machine-readable, detectable way. That duty sits with the tool's provider and is not observable from a public page. Article50.io reads image metadata for AI markers, but a missing marker is never reported as a finding.

Why might a questionnaire miss problems on my site?

A questionnaire knows what you tell it. It can't see a chat widget nobody remembers adding, a vendor default with no AI notice, or a disclosure that is hidden, tiny or unreadable by screen readers. A scan of the live page looks at what visitors actually see when it loads.

How long does an Article50.io scan take?

Each scan loads one public page in a real browser and is capped at 45 seconds. There is no signup for the free scan, which shows the single most severe finding and lists what it couldn't check. The €499 scan report covers every finding from the same scan.

When does Article 50 apply?

Article 50 applies from 2 August 2026 and was not delayed by the Digital Omnibus. Providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) marking duty, under Article 111(4), inserted by the Omnibus.

Was Article 50 delayed?

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. The full scan report is €499 one-off, with a ready-to-paste fix for every finding: see pricing.