Anthropic has announced that Claude models will include machine-readable markings in AI-generated content to comply with the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content.
These watermarks apply to output from supported models wherever Claude is available, including the Claude Platform (API), Claude Code, Claude Cowork, Claude Tag, and the main Claude experience.
Claude models launched in the EU on or after August 2, 2026 already include these markings, and Anthropic is working to extend support to models released before that date. The company also plans to publish technical documentation to help users and third parties detect Claude’s watermarks.
How the watermark works
According to the documentation, Claude watermarks content in two ways, depending on the content type. For generated text, an imperceptible watermark is embedded directly into the output. The mark travels with the text when it’s copied or pasted elsewhere but doesn’t change the meaning or quality of the content.
For generated files, including .svg, .png, and .jpg formats, Claude attaches digitally signed provenance metadata following the Coalition for Content Provenance and Authenticity (C2PA) open standard.
These markings apply worldwide, not just in the EU.
Limitations of the watermark
A detected mark signals that Claude may have processed content, but it doesn’t prove conclusively that Claude authored it. Content can carry a mark even when Claude was only used to proofread, translate, or summarize material that originated elsewhere.
At the same time, the absence of a mark doesn't mean the content wasn't AI-generated or AI-assisted. A mark may be undetectable if:
- The content was generated before marking support was available
- The text was heavily edited or paraphrased afterward
- The passage is very short
- File metadata was stripped through format conversion or re-saving
What content teams should know
Claude's markings give content teams a new signal to watch as search and AI platforms decide how to identify and treat AI-assisted content. As Chris Long pointed out on LinkedIn, machine-readable watermarks could make it easier for platforms like Google to detect content that Claude has processed. If platforms begin acting on those marks, the implications for content performance could be significant.
Answer engines and search engines have not announced whether they'll adopt this standard. But the infrastructure now exists for them to do so. Whether the outcome is a simple disclosure label on search engine results pages (SERPs) or a more consequential visibility adjustment for content carrying Claude marks, brands that rely on Claude in their workflows must plan for their response.
For content teams, the question is no longer whether AI-generated content can be detected. It's what platforms will choose to do with that information.
Next steps: Establish a benchmark for AI-assisted content
The most useful thing content teams can do right now is understand how their AI-assisted content performs. Establishing a baseline now makes it easier to identify changes if search and AI platforms ever act on these markings.
Semrush's AI Visibility Toolkit tracks citation frequency and visibility across ChatGPT, Perplexity, Google AI Mode, and Bing, so you can see which URLs are being cited, which queries are triggering those citations, and how that picture changes over time.
Position Tracking monitors daily keyword rankings in traditional search, which helps you measure any gap that emerges between SERP visibility and AI citation presence for the same queries.
Together, those data streams provide a baseline for understanding how your content performs today, while also making it easier to spot changes if platforms begin using Claude's markings as a signal in the future.
