Last verified: 30 October 2026 · Prices as checked 30 Oct 2026 · Not legal advice · Article50.io is our product, so we have an interest (see How we checked)
Yes. Several free tools check a website against Article 50 of the EU AI Act, but each sees something different: a questionnaire (Disclo), page source (EUAiShield), the rendered page (Article50.io, GotAlt, DisclosureProof) or the chatbot's replies (AnswerTested). No free tool proves compliance, so pick by the duty you need checked.
Key point: Article 50 has applied since 2 August 2026. The 2 December 2026 date is only a transition for Art. 50(2) machine-readable marking on generative AI systems already on the market before 2 August 2026. It is not "the Article 50 deadline".
Short answer: the best free Article 50 scanner depends on what you need checked
The best free Article 50 scanner is the one whose method can see the duty you need checked.
- Chat, voice or avatar widget shows a visible disclosure (Art. 50(1), 50(5))? Use a rendered-page tool.
- What the chatbot actually says when asked "am I talking to a bot?" (Art. 50(1))? Use a conversation test.
- A dated read of your page source? Use an HTML scanner.
- Not sure Article 50 applies to you at all? Start with a questionnaire, then scan.
Best for, as checked 30 Oct 2026 (use-case fit by method, not a score):
- Best free scanner for voice agents and AI avatars on public pages: Article50.io (rendered page, server-side browser).
- Best free check for staging or logged-in pages: GotAlt's AI label checker (bookmarklet in your own browser).
- Best free check of what a chatbot says: AnswerTested (conversation test).
- Best free dated, rule-versioned source report: EUAiShield (HTML observations).
- Best free scope questionnaire: Disclo (Scope Check).
The four ways a free tool can check Article 50
Free Article 50 tools use one of four methods, and each method has blind spots the others cover.
Questionnaire (self-assessment): sees only what you tell it
You answer questions about your AI use. It's useful for scope and documentation. It doesn't look at your site, so it can't catch a label that's missing or hidden.
HTML crawl: sees the source, may miss widgets injected by scripts
The tool fetches your page's HTML, which is fast and repeatable. Many chat widgets are added by JavaScript after the page loads, so a crawl that doesn't run scripts can miss them and their labels.
Rendered page (real browser): sees loaded widgets and how the label looks
The tool loads the page with scripts running, so it sees the widget a visitor sees and can check whether a disclosure is visible and readable. Some run a browser on a server (Article50.io, DisclosureProof); GotAlt runs in your own tab, so it can reach staging and logged-in pages but checks one page per click. A rendered scan generally can't see the conversation after it starts.
Conversation test: sees what the bot says
The tool sends questions to your chatbot and records the answers. It shows whether the bot tells people it's an AI. It can't see page labels or images.
Key point: each method has a blind spot. A rendered scan plus a conversation test covers more of Art. 50(1) and 50(5) than either alone.
[CTA] Run the free Article 50 scan — no signup, public pages, a scan is capped at 45 seconds. Findings describe what was seen; they are not a compliance certification or legal advice.
Which Article 50 duties a website scan can and can't check
A website scan can check Art. 50(1) disclosures, 50(4) labels and 50(5) presentation; it can't see 50(2) marking systems or 50(3) emotion recognition.
- Art. 50(1) — chatbots, voice agents, AI avatars: people must be told they are interacting with an AI system. A scan can check whether a disclosure is shown at the widget; a conversation test can check what the bot says.
- Art. 50(2) — machine-readable marking: a provider duty. A scan sees only metadata on published files (for example IPTC digital source type or C2PA Content Credentials), not the provider's marking system.
- Art. 50(3) — emotion recognition and biometric categorisation: generally can't be seen by a website scan.
- Art. 50(4) — deepfakes and AI-generated text on matters of public interest: a scan can look for a visible label next to the content.
- Art. 50(5) — how the information is given: clear, distinguishable, accessible, at the latest at first interaction or exposure. A rendered scan can measure presentation; a source-only scan sees less.
Why it matters: breaches of Article 50 can be fined up to €15 million or 3% of worldwide annual turnover, whichever is higher (the lower of the two for SMEs) — Art. 99(4)(g) and 99(6) (AI Act Service Desk, Article 99).
Free Article 50 scanners compared (checked 30 Oct 2026)
As checked on 30 October 2026, at least eight tools offer a free Article 50 check, including the eight below. They differ in method, duties covered and evidence. For paid plans, evidence packs and monitoring tools, see Article 50 website checkers compared.
Key: "✓" = the tool's own page says it checks this; "partial" = checks part of the duty; "self-reported" = from your answers, not observed (Disclo); "not observable" = the tool says a public page can't show it; "not stated" = not on its page (not the same as "can't"). Prices as checked 30 Oct 2026.
| Tool | Method | Free tier | Paid price (30 Oct 2026) | 50(1) | 50(2) | 50(4) | 50(5) | Languages | Evidence output | Page last dated | Stated limit |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Article50.io (ours) | Rendered (Chromium, server-side, scripts run) | Free scan, no signup, public pages; shows the most severe finding | €499 full report, one-off | ✓ chat, voice, avatar widgets (not what a voice agent says on a call) | partial: reads IPTC/C2PA AI markers on up to 8 sampled images, but only to flag a missing visible label; the marking duty itself is "not observable from a public page" (optional answer shown as "Declared by you, not verified") | partial: images whose file data says AI-generated, without a visible label | ✓ visibility, size, contrast, accessibility | Detects disclosure wording in EN, DE, FR, ES, IT, NL, PT; image/content labels English only | Dated (timestamped) PDF report with a unique report ID and SHA-256 fingerprint, copy-paste text, implementation steps, "Checked" badge, one free recheck within 30 days | Homepage 27 Oct 2026; how-the-scan-works 7 Oct 2026 | Doesn't see the conversation after it starts |
| Article 50 Ready | Platform detection (35+ chat/AI-widget platforms) | Free scanner | €29 one-time, single site ( | ✓ detects 35+ platforms | not stated | not stated | not stated | Generates widget notices in 24 EU languages (detection languages not stated) | Evidence PDF | No date on page | "Our scan flags what we can see" |
| Disclo | Questionnaire (7-question Scope Check); not a site scanner | Free Scope Check | Kit €69 one-time; Pro €149/yr excl. VAT | self-reported | self-reported | self-reported | not stated | Site/texts in 6 EU languages | Evidence log (render log, PDF/CSV) | "Last content review: 12 July 2026" | "The evidence log… does not by itself establish legal compliance" |
| EUAiShield | HTML observations; "The scanner does not execute JavaScript" | Free public scans, no account | not stated | partial: script signatures and notice excerpts in HTML | partial: sampled image metadata (up to 5 images); no C2PA signature validation | partial: flags a missing text notice after you confirm applicability | not stated (CSS visibility "requires manual review") | Site UI EN/DE/FR/ES; detection languages not stated | Dated reports, SHA-256 source fingerprint, rule version, PDF/JSON | Rules reviewed 1 Oct 2026 | "Technical observations support your review. They do not certify legal compliance or verify C2PA signatures." |
| GotAlt AI label checker | Bookmarklet; reads the rendered page in your browser | Free, no signup, unlimited | Site Image Check $49 (≤50 pages) / $149 (≤250 pages), one-off; AI disclosure monitoring from $69/mo (Professional); USD excl. tax | partial: 35 chat-widget signatures plus visible launcher wording; doesn't open widgets | not stated | ✓ AI label wording near images, alt text | ✓ text vs pixels, accessible name, aria-hidden, hover-only, WCAG contrast, position | EN, DE, ES, FR, IT, PT, NL | On-screen panel (no export seen) | Published and last reviewed 21 Oct 2026 | "It sees one page, once"; "never clicks, opens, scrolls or types"; "Some sites can block it" |
| AnswerTested | Conversation test | $0: one test of one chatbot, 10 questions, no card | Starter $29/month (founding $17.40/mo for 12 months, first 20 paying customers); USD before tax | ✓ what the bot says at start and when asked directly | not stated | not stated | not stated | not stated | Printable page plus SHA-256-hashed JSON; CSV/PDF; no screenshots | "Checked 21 October 2026" | "It records what your chatbot said on the day we tested it." |
| DisclosureProof | Rendered (desktop + mobile) | "The free scan checks your homepage"; no account; 90-second capture limit; six core findings | Evidence Pack €79 (sealed PDF kept 365 days, SHA-256/HMAC) | ✓ disclosure at first interaction (free); mobile persistence (not stated if free) | ✓ (paid advanced): IPTC/XMP and structural C2PA in sampled media, no signature validation | ✓ (paid multi-page scans): AI-content label on article-like pages | partial: first interaction and mobile persistence | "Declared-language coverage" (list not stated) | Free: available captured evidence; paid: sealed Evidence Pack | No page date seen; rule pack 2026.12 | "we never call a site compliant" |
| Shieldra | Single fetch of the page you paste; looks for AI-interaction disclosure phrases (part of a broader compliance platform) | Free ("Free AI Disclosure Scanner") | not stated | ✓ disclosure phrases only | not stated | not stated | not stated | English and "the major EU languages" (list not stated) | Disclosure verdict | No page date seen | "A disclosure verdict is one piece of Article 50" |
Key point: language counts measure different things. Article50.io and GotAlt state the languages they detect disclosure wording in (7 each); Article 50 Ready states the languages its widget generates notices in (24). Compare like with like.
Article50.io (our product)
- Method: loads the public page in Chromium on our server with scripts running, and reads text inside up to 25 chat-like frames. Recognises 15 vendor signatures — chat (9): Intercom, Drift, Zendesk, Crisp, Tidio, LiveChat, HubSpot chat, Freshchat, tawk.to; voice (3): ElevenLabs, Vapi, Telnyx; avatar (3): D-ID, Tavus, Synthesia. A scan is capped at 45 seconds.
- Best at: Art. 50(1) chat, voice and avatar widgets on public pages; Art. 50(5) presentation (visibility, size, contrast, accessibility); images whose file data says AI-generated but show no visible label. Detects disclosure wording in 7 languages (EN, DE, FR, ES, IT, NL, PT).
- Stated limit: doesn't see the conversation after it starts or what a voice agent says on a call; content and image labels are detected in English only; it reads IPTC and C2PA markers on sampled images but can't check whether you apply machine-readable marking (Art. 50(2)); the report lists that as not observable, with any questionnaire answer shown as "Declared by you, not verified".
Article 50 Ready
- Method: free scanner that detects 35+ chat/AI-widget platforms.
- Best at: wide platform detection; generates disclosure notices in 24 EU languages; evidence PDF. €29 one-time for a single site (€49 struck through), €119 for 10 sites, labelled a launch price.
- Stated limit: "Our scan flags what we can see." As of 30 Oct its homepage refers to "draft guidelines" and says it "applies from 2 August 2026"; the Commission guidelines were finalised on 20 July 2026.
Disclo
- Method: a 7-question Scope Check — a questionnaire, not a site scanner.
- Best at: scoping and documentation, with an evidence log for your records. Kit €69 one-time; Pro €149/yr excl. VAT.
- Stated limit: "The evidence log… does not by itself establish legal compliance." As of 30 Oct the page shows "Last content review: 12 July 2026", a banner saying final guidance is expected, and a countdown at 0 days "until Article 50 applies".
EUAiShield
- Method: HTML observations from free public scans, no account needed. Its methodology page says: "The scanner does not execute JavaScript, open conversations, inspect iframe contents…"
- Best at: dated, reproducible reports with a SHA-256 source fingerprint and rule version, as PDF/JSON.
- Stated limit: "Technical observations support your review. They do not certify legal compliance or verify C2PA signatures." Because it doesn't run JavaScript, it may miss widgets that scripts inject.
GotAlt AI label checker
- Method: a free bookmarklet that reads the page as your browser has rendered it, so it works on staging, localhost and logged-in pages.
- Best at: Art. 50(5) presentation — real text vs pixels, accessible name, aria-hidden, hover-only, WCAG contrast, position — plus label wording in 7 languages and 35 chat-widget signatures.
- Stated limit: "It sees one page, once"; it "never clicks, opens, scrolls or types"; "Some sites can block it"; and "It does not say whether a law applies or is met."
AnswerTested
- Method: sends questions to your chatbot and records the answers.
- Best at: what the bot actually says at the start and when asked directly; printable evidence plus SHA-256-hashed JSON.
- Stated limit: "It records what your chatbot said on the day we tested it." Our inference: it tests the conversation, not page labels or images.
DisclosureProof
- Method: rendered capture on desktop and mobile; "The free scan checks your homepage", with six core findings and no account needed. (Not to be confused with Disclo.)
- Best at: the free homepage scan checks disclosure at first interaction on desktop and mobile; AI-content labels on article-like pages are checked only in paid multi-page scans, and its page doesn't state whether mobile persistence is free. The paid Evidence Pack (€79) is a sealed PDF kept 365 days with SHA-256/HMAC.
- Stated limit: "The free scan checks your homepage"; "we never call a site compliant".
Shieldra
- Method: a single fetch of the page you paste, looking for AI-interaction disclosure phrases in English and "the major EU languages"; part of a broader compliance platform.
- Best at: a quick Art. 50(1) check for disclosure wording on one page.
- Stated limit: "A disclosure verdict is one piece of Article 50."
What the Commission guidelines say a good disclosure looks like
The Commission's Guidelines on transparency obligations (C(2026) 5054 final, 20 July 2026) are final and non-binding, and they set a practical bar for disclosures:
- First turn, with persistent labels as appropriate (¶37): first-turn greetings "may be combined, as appropriate, with persistent badges or labels visible throughout the interaction".
- Generic notices aren't enough (¶38): "a general disclosure like 'Services on this website use AI' is insufficient", and a disclosure only in terms and conditions isn't enough on its own.
- Clear means noticeable (¶142): "clear… where it is noticeable, easy to understand… accessible".
- Format and vulnerable groups (¶34).
- Accessibility (¶144): Article 50 "does not impose distinct or additional accessibility requirements"; existing accessibility rules apply where they bind you.
Key point: a footer line saying "Powered by AI" is unlikely to do the job on its own. Put the disclosure at the chat itself, from the first message.
Can a free scanner prove compliance?
No. A scan reports what it saw on one date. It can't see your contracts, your provider's marking system, or every conversation. What it can give you is evidence: dated reports, render logs, hashed JSON exports (where a tool offers them). Keep them with your own documentation, and ask a lawyer about your case.
How to combine free tools
A rendered-page scan plus a conversation test covers more of Art. 50(1) and 50(5) than any single free tool.
- Rendered scan — check that the widget shows a visible, readable disclosure at first view (50(1), 50(5)).
- Conversation test — check the bot says it is an AI when asked (50(1)).
- Evidence — keep dated reports from each tool, and recheck after site changes.
For context: in our study of home pages scanned 21–22 Oct 2026, 58 of 59 (98.3%) popular EU websites with a chat widget showed no AI disclosure detected by an automated scan at first view, before the chat was opened (State of Article 50 on EU websites 2026). Some may disclose once the chat opens.
[CTA] Run the free scan (no signup, public pages, capped at 45 seconds). Need copy-paste fixes and a downloadable report? The full report is €499, one-off, with a free recheck within 30 days. Findings describe what was seen, not a compliance certification or legal advice.
How we checked
On 30 October 2026 (NZDT) we opened each tool's live public pages and recorded method, free tier, price, duties covered and stated limits. We list only facts we could see that day; anything else is marked "not stated", and our own inferences are labelled. Article50.io is our product, so we have an interest; we don't rate or score rival tools. See how the scan works, the Article 50 guide, and our broader Article 50 website checkers compared, which includes paid tools.
Frequently asked questions
Is there a free Article 50 scanner?
Yes. As checked on 30 October 2026, at least eight tools offer a free Article 50 check, including Article50.io, Article 50 Ready, Disclo, EUAiShield, GotAlt's AI label checker, AnswerTested, DisclosureProof and Shieldra. They use different methods, so pick the one that sees the duty you need checked.
Can a free scanner prove I comply with Article 50?
No. A free Article 50 scanner shows what it saw on one page on one date. That is useful evidence, not a certificate. Compliance depends on facts no scan can see, such as your provider's marking system or every conversation. This is not legal advice.
Does a scanner check what my chatbot actually says?
Most page scanners don't. Article50.io and GotAlt read the page and widget before the chat is opened, and DisclosureProof checks disclosure at first interaction. A conversation test such as AnswerTested sends questions to the chatbot and records its answers, so it checks what the bot says.
Is 2 December 2026 the Article 50 deadline?
No. Article 50 of the EU AI Act has applied since 2 August 2026. The 2 December 2026 date is only a transition for Art. 50(2) machine-readable marking on generative AI systems that were already on the market before 2 August 2026. Other Article 50 duties already apply.
Does a "Powered by AI" footer count as a disclosure?
On its own, probably not. The Commission's final Article 50 guidelines (¶38) say a general disclosure like "Services on this website use AI" is insufficient, as is a notice only in terms and conditions. Put a clear AI disclosure at the chat itself, from the first message.
Can a scanner check AI-generated images?
Partly. A scanner can read image metadata such as IPTC digital source type or C2PA Content Credentials, and look for a visible AI label nearby. It can't confirm how the image was made. Article50.io flags images whose file data says AI-generated but that show no visible label.
Which free scanner checks voice agents and avatars?
Article50.io's free scan checks chat, voice and AI avatar widgets on public pages, loaded in a real browser. It recognises voice widgets from ElevenLabs, Vapi and Telnyx and avatars from D-ID, Tavus and Synthesia. It doesn't check what a voice agent says on a call.
What is in Article50.io's €499 report?
Article50.io's full report costs €499, one-off. It includes a dated (timestamped) PDF report with a unique report ID and a SHA-256 fingerprint, copy-paste disclosure text, implementation steps, a "Checked" badge and one free recheck within 30 days. Findings describe what the scan saw; the report is not a compliance certification or legal advice.
Sources
- Regulation (EU) 2024/1689 (EU AI Act), Article 50 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
- European Commission AI Act Service Desk, Article 99 — https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99
- European Commission, Guidelines on transparency obligations, C(2026) 5054 final, 20 Jul 2026 — https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems · PDF: https://ai-act-service-desk.ec.europa.eu/sites/default/files/2026-07/guidelines_on_the_implementation_of_the_transparency_obligations_for_certain_ai_systems_under_article_50_of_the_ai_act_bzptwqhk0ikg1dtlddap41psfy_131215.pdf
- Article50.io study — https://article50.io/research/state-of-article-50-eu-websites-2026 · method: https://article50.io/how-the-scan-works
- Tool pages, checked 30 Oct 2026: https://article50ready.com/ · https://disclo.eu/ · https://euaishield.com/ (methodology: https://euaishield.com/methodology/) · https://gotalt.com/tools/ai-label-checker/ · https://gotalt.com/pricing/ · https://answertested.com/eu-ai-act · https://answertested.com/pricing · https://disclosureproof.com/ · https://www.shieldra.ai/ai-disclosure-scan
Not legal advice. Findings and prices as checked 30 October 2026.