AI Content Authenticity Checker
Upload an image or paste text to find out if it carries AI generation markers. We read C2PA signed provenance in images from Claude, DALL·E, Firefly and others, and scan text for hidden Unicode characters. Free, no account needed.
or
Files are analysed in your browser session, not stored. Text files get the full scan; PDF/DOCX get a metadata/provenance scan.
What this checks and what it doesn't. Images: we read C2PA and IPTC provenance, extract the generator name, and identify files from Claude, DALL·E, Firefly and others. Text: we detect invisible and zero-width Unicode characters, estimate text length adequacy, and estimate how much of the text was "markable" by a watermark scheme.
Statistical text watermarks are a different class entirely. Three schemes are in production: Kirchenbauer et al.'s KGW (used by Claude, Aug 2026+), Google's SynthID-Text (Gemini), and Aaronson's scheme (reported in ChatGPT). All three work by secretly tilting word choices at generation time. The mark lives in patterns of decisions, not in any characters. Detecting them requires the provider's private key. Your teacher, editor, or any third-party "AI detector" website cannot run this test. We have the integration seam built; it goes live the day Anthropic opens the API.
A found mark means "processed by", not "written by." Anthropic's own documentation notes that human text Claude merely proofread or translated picks up the mark. Absence proves even less: old models or heavy editing yield clean results on genuine AI text. We show you only what is machine-verifiable.
Statistical text watermarks are a different class entirely. Three schemes are in production: Kirchenbauer et al.'s KGW (used by Claude, Aug 2026+), Google's SynthID-Text (Gemini), and Aaronson's scheme (reported in ChatGPT). All three work by secretly tilting word choices at generation time. The mark lives in patterns of decisions, not in any characters. Detecting them requires the provider's private key. Your teacher, editor, or any third-party "AI detector" website cannot run this test. We have the integration seam built; it goes live the day Anthropic opens the API.
A found mark means "processed by", not "written by." Anthropic's own documentation notes that human text Claude merely proofread or translated picks up the mark. Absence proves even less: old models or heavy editing yield clean results on genuine AI text. We show you only what is machine-verifiable.
Want to track how AI engines cite and represent your brand?
Citerank monitors your AI citations across ChatGPT, Gemini, Perplexity and more, and shows you what to fix to earn more of them.
Explore Citerank free →