Claude now marks all AI-generated content with invisible watermarks and signed metadata. If you use Claude for content, your audience can detect it.
What Changed
Anthropic signed the EU AI Act’s Article 50(2) Code of Practice. Starting August 2, 2026, all new Claude models mark content at launch:
- Invisible watermarks embedded in text — travel with copy/paste, persist through some editing
- C2PA metadata attached to files (.svg, .png, .jpg) — follows open standard for content provenance
- Works everywhere — Claude API, Claude.ai, Claude Code, AWS, Google Cloud, Microsoft Foundry
Why This Matters for SEO
Google’s helpful content system penalises low-quality AI content. Watermarked content makes it easier to identify:
- Search engines can potentially detect watermarked content
- Readers can verify if content was AI-generated
- Clients may ask if your content is AI-assisted
What You Can Do Now
If you use Claude for content:
- Review output before publishing — add your own voice and expertise
- Don’t rely on AI for E-E-A-T signals (experience, expertise, authoritativeness, trustworthiness)
- Mix AI drafts with original research, data, and case studies
If you create content for clients:
- Disclose AI assistance if asked
- Ensure human oversight on all published content
- Focus on unique insights AI can’t generate
What’s Limitied
Watermarks aren’t foolproof:
- Heavy editing can remove or degrade the watermark
- Short passages may not carry a detectable signal
- File metadata can be stripped through format conversion
- Pre-August 2026 Claude models don’t mark content yet
Source: Anthropic — How Claude Marks AI-Generated Content
Related: Optimise for AI Search → | GSC Platform Properties →