DisclosureProof · Article 50 disclosure checks
Check your site’s
AI disclosures.
We inspect visible AI disclosures and the provenance markings inside sampled media. Six core findings are free, with the available captured evidence.
The free scan checks your homepage, without an account or card, within a 90-second capture limit.
We are not a law firm, and we never call a site compliant.
Read the
whole surface.
We capture the homepage at two viewport sizes and inspect its markup. Chat-panel text is one part of the record.
Observed on the page
- Site language markup
- en / de
- AI policy link
- /ai-policy
- Sampled image files
- 2
- Chat widget
- intercom
Hi there! How can we help you today?Mobile
Hi there!
- Desktop capture setting
- 1366 × 768
- Mobile capture setting
- 390 × 844
Screenshots are included when capture succeeds. A widget we cannot open has no chat-panel screenshot. Direct questions require a signed-in account with verified domain ownership.
- Free scan scope
- 1 homepage · 90-second capture limit
- Additional checks
- Mobile persistence and declared-language coverage
Inside
the file.
IPTC and XMP markings. The advanced check inspects structural C2PA data in sampled files.
hero-illustration.pngSampled file| Sampled file | Marking detected |
|---|---|
hero-illustration.png | XMP |
blog-cover.jpg | Not detected |
We read the first 512 KB of each sampled asset, within fetch limits. The absence of marking does not establish that a file was AI-generated. C2PA structure is inspected without cryptographic signature validation.
- In this fixture
- 1 of 2 files carry a marking
- Structural manifests
- 0 detected in this fixture
- Advanced finding
- ▲ Attention
Across
the site.
in this fixture
We compare article-like pages with AI-content labels in their markup and publisher-emitted signals. The report identifies the pages where a label was not detected.
| Page path | Page type | AI-content label |
|---|---|---|
/rendered | Other page | N/A |
/blog/ai-in-logisticsstatic | Article-like | Not detected |
/blog/q3-market-outlookstatic | Article-like | Not detected |
/news/expansion-2026static | Article-like | Not detected |
/blog/how-we-writestatic | Article-like | Detected: ai-assisted |
/aboutstatic | Other page | N/A |
- Article-like pages
- 4
- With a label
- 1
- Label not detected
- 3
This fixture’s AI-use signal is uncorroborated; manual verification is recommended. We do not use statistical AI-text detection or determine that a page was written by AI.
- Free
- 1 homepage
- Verified Pro / Business sites
- Homepage + up to 10 / up to 500 pages
- Evidence Pack purchase
- Separate crawl report · up to 100 pages
9 checks.
One report.
Six core checks and three advanced checks. Each finding keeps its own state, including the checks that require your declaration.
| Reference | Check | State |
|---|---|---|
| Art. 50(1) | Chatbot AI disclosure at first interaction | ▲ Attention |
| Art. 50(1) | Chatbot disclosure persists on mobile | – N/A |
| Art. 50(1) | Chatbot disclosure covers the site's languages | – N/A |
| Art. 50(2) | Machine-readable provenance marking on media | ▲ Attention |
| Art. 50(2) (supporting) | Provenance marking is structural, not textual-only | ▲ Attention |
| Art. 50(4) | AI-content label on article-like pages | ▲ Attention |
| Art. 50(4) (supporting) | Official EU labelling icons in use | – N/A |
| Art. 50 (supporting) | AI policy / disclosure page | ● Pass |
| Art. 50(3) | Emotion recognition / biometric categorisation disclosure | – N/A |
Inside a finding · 3 affected pages · 4 fix steps
Illustrative paid detail · fixture.disclosureproof.com
AI-content label on article-like pages
Root cause: one publisher-emitted signal indicates AI tooling is in use here, uncorroborated by any other kind of evidence, and 3 of 4 article-like page(s) carry no AI disclosure. A single signal does not establish which pages a tool produced. Confirm which content the tool was used for.
Recorded page paths
/blog/ai-in-logistics/blog/q3-market-outlook/news/expansion-2026
Fix guidance
- If any published text is AI-generated on a matter of public interest, add a visible label (the Code of Practice's uniform wording/icons are the safe default)
- Human-reviewed content with editorial responsibility can rely on the Art. 50(4) carve-out — document the review process
- Add the label near the byline or article footer where a reader actually sees it, not only in metadata
- State your AI-use policy on a dedicated page and link it site-wide
- Free findings
- 6 core
- Paid report layer
- 3 advanced checks
- Reproducible rules
- 2026.12
Paid reports add recorded page and asset details, with root-cause explanations and fix guidance where available. Findings remain separate; a state is not a whole-site verdict.
Four Article 50 duties and their dates
Four duties.
Article 50 has applied since 2 August 2026. The existing-system marking date is 2 December 2026.
| Duty | Surface | Applies |
|---|---|---|
50(1) | Chatbots must say they’re AIClear disclosure at the first interaction, unless it is already obvious to a reasonably well-informed person. | 2 Aug 2026 |
50(2) | Machine-readable markingAI-generated audio, image, video and text. Existing generative systems have the later marking date. | 2 Aug 2026 2 Dec 2026* |
50(3) | Emotion recognition and biometric categorisationTell the people exposed to it. | 2 Aug 2026 |
50(4) | Deepfakes and public-interest textVisible disclosure of AI generation or manipulation. | 2 Aug 2026 |
* The 2 December 2026 date concerns 50(2) marking for generative systems already on the market before 2 August 2026.
Read the four duties and their limits ↗- Coverage
- 9 checks = 6 core + 3 advanced
- Outside observation
- 2 checks require your declaration
- Rule pack shown below
- 2026.12
Our published sweep · historical results and method correction
78%of confirmed widgets were unreadable to our automated visitor.
Historical measurement. The 78% unreadable figure was measured with the broken consent routine. We have not measured how much this bug changed that figure. Method correction — 3 October 2026 ↗
Each square shows the furthest stage reached. The totals below are cumulative; the key runs from a site swept to a disclosure read.
- 1,088sites swept
- 794widgets confirmed
- 259panels opened
- 174first messages readable
- 19of those 174 disclosed AI
A single post-application snapshot. The read-only sweep typed no questions into third-party chatbots. It publishes no compliance verdicts.
- Scan date
- 10 August 2026
- Study rule pack
- 2026.07.9
- Research
- Results and exclusions ↗
The €79 Evidence Pack preserves this scan’s sealed PDF for 365 days. Its included crawl produces a separate report of up to 100 pages.
Keep the
record.
File integrity
Anyone can re-check the SHA-256 hashes against the captured files.
Independent calculationOur signature
The seal is an HMAC. Confirming that it is ours relies on DisclosureProof.
DisclosureProof confirmationRecorded time
From DisclosureProof’s clock. It does not provide an independent trusted timestamp.
Clock and capture limits statedSHA-256 of the pipeline fixture projection
50a672387dad192c7d00f2c2ae526eaeec350f8496d914422d439d3b605fe94fThis file contains the HTML preview’s fixture findings. Its hash checks those file bytes; it is not an Evidence Pack seal. To try the integrity check, download the file and calculate its SHA-256 locally, then compare the result with the 64-character value above.
- Hash algorithm
- SHA-256
- Signature algorithm
- HMAC-SHA256
- Verification
- Open the verification tool ↗
Catch
a change.
Each scheduled scan is compared with the previous record. Changes can include a lost media marking or an article label as well as a chat disclosure.
| Signal | Previous | Current |
|---|---|---|
| Chatbot AI disclosure at first interaction | ● Pass | ▲ Attention |
| Chatbot disclosure on mobile | ● Pass | – N/A |
| Chatbot disclosure language coverage | ▲ Attention | – N/A |
| Machine-readable provenance marking on media | ● Pass | ▲ Attention |
| AI-content label on article-like pages | ● Pass | ▲ Attention |
| Official EU labelling icons | ▲ Attention | – N/A |
| Chat widget disclosure changed | discloses AI (ai assistant) on panel | no AI disclosure detected |
| Media provenance-marking ratio dropped | 100% marked | 50% marked |
fixture.disclosureproof.com: 8 Article 50 changes detected (review suggested)
- Pro
- 1 monitored site · weekly
- Business
- 10 monitored sites · daily
- Monitoring prerequisite
- Signed-in subscription · verified domain ownership
Start with
the findings.
- Free scan 1 homepage · 6 core checks
- €0
- Evidence Pack Per scan · sealed PDF · separate crawl up to 100 pages
- €79 once
- Pro Homepage + up to 10 key pages · 1 monitored site weekly
- €20 / month
- Business Up to 500 pages per scan · 10 monitored sites daily
- €200 / month
Pro and Business scan depth and monitoring require verified site ownership. Eligible first-time subscriptions have a 7-day trial; the Evidence Pack is a one-time purchase.
Plan limits, retention and annual prices ↗- Evidence Pack retention
- 365 days
- Pro retention
- 90 days
- Business retention
- Subscription-lifetime
Before the scan
What does the free scan include?
One homepage, six core findings and the available captured evidence. Browser capture is capped at 90 seconds. Checks with no applicable signal, unreadable surfaces or required declarations remain explicitly marked in the report.
How does a multi-page scan work?
Verified Pro sites include the homepage plus up to 10 key pages; Business sites can be scanned to 500 pages. Additional pages are fetched as HTML, while widget findings remain based on the homepage interaction. An Evidence Pack purchase includes a separate crawl report of up to 100 pages; its sign-in and confirmed-email requirements are listed on the pricing page.
Can you determine whether content was generated by AI?
We inspect embedded media markings and publisher-emitted signals, including labels in page markup. We do not use statistical AI-text detection or determine whether an image is a deepfake. Missing metadata alone does not establish AI generation; C2PA structure is inspected without cryptographic signature validation.
Will the scan send a message to my chatbot?
Direct chatbot questions require a signed-in account with verified domain ownership. The anonymous free scan tries to read a supported widget without sending a question. On verified sites the scan also types a short test message into your chat widget to see how it answers, which can create a conversation in your support inbox.
What can I independently re-check?
Anyone can calculate the SHA-256 hashes against the captured files. Confirming the HMAC seal relies on DisclosureProof. Recorded time comes from DisclosureProof’s clock, and the seal does not independently prove authenticity or capture time.
Is a DisclosureProof scan legal advice or a compliance guarantee?
No. DisclosureProof is an informational scanning and evidence tool, not a law firm, and a scan is not a certification or a guarantee of compliance. It is built to surface disclosure gaps and preserve dated proof of what your site displayed; for how the law applies to your specific circumstances, use qualified counsel.
Your site,
on record.
Begin with six core findings on one homepage, without an account or card. Capture is capped at 90 seconds.
- Free checks
- 6 core
- Scope
- 1 homepage
- Direct questions
- Verified domain ownership required