Checker › Guides › AI Ad Disclosure Rules
AI ad disclosure rules, explained by jurisdiction
Three rule sets now decide whether an AI-made ad is compliant: the EU's AI Act, Google's ad-platform rules, and China's labeling measures. They agree on the shape — tell people, and mark the file — and differ on the details. This guide walks each one, then shows what real brands' creatives actually score. To test your own creative as you read, use the free disclosure checker.
European Union — AI Act Article 50
Article 50 is the transparency article. Two duties matter for advertisers. First, machine-readable marking: outputs of AI systems must be marked in a format machines can detect, where technically feasible — in practice, provenance metadata such as a C2PA content-credentials manifest or an XMP packet naming the generator. Second, visible disclosure: people exposed to AI-generated or manipulated content must be told, clearly and at first exposure. The strictest case is synthetic depictions of real people — deepfakes — where the disclosure must be unmistakable. In the checker on this site, a real-person asset with no visible label is the only CRITICAL finding in the EU pack (30 points).
United States — Google Ads synthetic-content rules
The US has no single federal AI-disclosure statute for ads; the operative rules are platform rules, and Google Ads is the one with reach. Google requires advertisers to disclose synthetic content in ads, provides the “How this ad was made” disclosure copy for the ad or its landing page, and expects a visible label on the creative where the format supports it. For a working checklist, Google's synthetic-content disclosure rules, explained step by step is the most complete public walkthrough we know. In this site's checker, the US pack tests exactly two things: the disclosure copy exists, and a visible label is attached.
China — AI-generated content labeling measures
China's measures split labeling into two halves that map neatly onto the file itself. The explicit label is visible to the audience — a notice on or around the content. The implicit label lives in the file's metadata — the provenance record. A third element, watermark guidance, asks distributors not to strip the generator's embedded watermarks. If your campaign never reaches Chinese platforms or audiences, this pack is optional; the moment distribution crosses over, both halves are expected.
What real creatives score — a measured sample
Rules are abstract until they meet a real file. In October 2026 we ran public ad creatives from seven consumer brands — wellness, beauty, and beverage names whose ads run heavily on social platforms — through this site's engine (disclosure rules v2026.10.2, EU + Google Ads + China packs). Brands are anonymized; the numbers are the engine's own output.
Source: SiteCredly research sample, 7 consumer brands' public ad creatives, scanned 26 October 2026, rules v2026.10.2. Method: the same engine as the checker on this page — declared provenance only (C2PA, XMP, EXIF, generator text) plus recorded label state; no AI-generation guessing.
Read that table again: the best-funded creative in the sample still failed. Not because the rules are exotic — because disclosure is a production step nobody had added to the pipeline. The gap is almost never the label itself; it's that the export step strips or never writes the metadata, so even a brand that intends to disclose can't prove it from the file.
What “good” looks like
The labeled sample on the checker page scores 90/A: an XMP packet naming the generator, plus disclosure copy a reader can see. Its only remaining finding is the upgrade from plain XMP to a signed C2PA manifest. That is the whole playbook: generate, keep the provenance at export, attach the visible label, add the platform copy. Four production habits, and all three jurisdictions are satisfied at once.
Check a creative now
Drop a creative here (JPEG, PNG, WebP, MP4) — or
The checker runs in your browser — your file never leaves your device. Full version with ad-copy scanning and samples: the checker hub.
Frequently asked questions
What is the EU AI Act Article 50 disclosure duty?
Article 50 requires providers and deployers of AI systems to tell people when content is AI-generated or manipulated. For advertising, that means a visible disclosure to the audience and machine-readable marking of the output where the format allows it.
What does Google Ads require for synthetic content?
Google Ads requires advertisers to disclose when an ad contains synthetic or AI-generated content, using the disclosure copy Google provides ('How this ad was made') and a visible label on the creative where the format supports it.
Do China's labeling measures apply to ads made outside China?
They apply to AI-generated content distributed to the public in China. If your campaign reaches Chinese platforms or audiences, the measures expect both an explicit visible label and an implicit metadata label.
What happens if a creative has no label and no metadata?
Under the EU pack in this site's checker, that creative scores 40 out of 100: three high-severity findings at 20 points each — no embedded provenance, no machine-readable marking, and no visible label.
A campaign's disclosure risk lives across every file, placement and landing page — not in a single upload box.
Get the full website scan Get a full disclosure auditRelated: C2PA, XMP, EXIF: where content provenance actually lives in a file — the metadata this guide's rules depend on, explained field by field.