Built to check the watermark layer—not guess authorship.
Check AI Watermarks is an independent browser-based tool for inspecting watermark signals that can be verified directly in text, image, and video files.
It turns supported machine-readable signals into inspectable evidence: exact Unicode characters, image metadata, and embedded image or video Content Credentials. The checker does not use an LLM, rewrite prose, score writing style, or alter media pixels.
Why the distinction matters
“AI watermark” can mean several incompatible technologies. Some marks are literal characters or file metadata. Others are statistical patterns in token or pixel choices and require a provider's secret key or detector. Treating all of them as one invisible tag produces misleading results. This project reports the layer it actually tested and explains what a zero does—and does not—mean.
How claims are verified
Implementation claims are tied to published standards, primary vendor documentation, versioned scanner rules, automated fixtures, and reproducible research. The methodology page lists the exact cleanup policy and validator versions. The 10,000-response study publishes its method and limitations instead of turning a sparse finding into an AI-authorship claim.
Editorial principles
- Say what was checked. A negative result applies only to supported layers.
- Separate structure, validity, and trust. Finding a C2PA container is not the same as validating its signature or trusting its signer.
- Preserve content by default. Ambiguous, malformed, language-sensitive, emoji-sensitive, and math-sensitive signals are retained for review.
- Prefer primary sources. Technical explanations link to the standard or provider responsible for the mechanism.
Independent, not provider-operated
Check AI Watermarks is not affiliated with or endorsed by Anthropic, Google, OpenAI, the Coalition for Content Provenance and Authenticity, or the Content Authenticity Initiative. Product names and standards are referenced only to explain compatibility and detection limits.