Anthropic is now watermarking Claude's writing

Every schoolkid discovers the same trick with lemon juice: write with it and the page looks blank until someone holds it to a flame. Claude's new watermark works on a similar principle: it's invisible on the page and detectable only to someone who knows where to look. 

Richard Stone, founder of technical PR agency Stone Junction, looks at what Anthropic's move to mark AI-generated text means for anyone using AI to help draft PR and marketing copy.

Claude models launched in the EU on or after August 2, 2026 embed a machine-readable watermark in every piece of generated text. It is automatically applied and is invisible to the reader. Files such as images carry signed provenance metadata using the C2PA standard, where the format supports it.

Anthropic is applying the change globally, rather than only in the EU. Article 50 of the EU AI Act requires deployers of AI systems that generate public-interest text to disclose that AI was used, with the duty falling on the publisher or content producer using the system. 

Anthropic has said the global rollout comes down to a technical limit, as it doesn't yet have a reliable way to apply the watermark only within the EU, so every Claude model gets it. Clearly, the requirement to comply, as a user, only exists if you are inside the EU or sharing your content inside the EU.

For models already in use before 2 August, Anthropic is adding watermarking support in stages over the coming months.

Anthropic says the change has no effect on output quality and that watermarked text reads identically to unwatermarked text. The method statistically biases the words Claude selects during generation, creating a pattern too subtle for a reader to notice but detectable by software built to look for it.

How much editing removes the Claude watermark?

A detected watermark shows that Claude was involved in producing a piece of text. It does not show how much of that text is still Claude's own drafting by the time it reaches a reader.

Anthropic's own guidance is very clear. Light editing will likely leave a watermark. A complete rewrite that replaces every word removes it.

Between those two extremes sits most PR and marketing copy produced with AI: a draft is generated, then added to, restructured and tightened before it reaches a client. It is rarely retyped from scratch.

Content that has been through that kind of editorial pass may still carry Anthropic's mark, even where a writer would reasonably describe the piece as their own. After all, AI produced its words from your prompt, which you probably wrote using the ask Traci prompt writing structure, and sweated blood over, right? 

Ryan Law, a content marketing director at Ahrefs who has written about the rollout, explains the real risk plainly: it only becomes a problem when “you're passing off AI work as human-written when you've promised otherwise.”

The editorial standard agencies now need

Many marketing teams and organisations have treated a human editorial pass as sufficient grounds to call a piece human-authored. That assumption gets harder to defend once a watermark can survive exactly that kind of editing.

Article 50 places the disclosure obligation on the organisation publishing the content, rather than the tool provider. Agencies are not the ones legally required to label copy under EU rules. Content only avoids the disclosure obligation once a named person has properly read it and taken responsibility for what is published, not just glanced at it. The Commission's guidance rules out spell-checking as a stand-in for real review.

There's a reputational risk, too. A client told that a piece was written by a human at their agency may question that claim if they then discover a heavily edited AI-generated draft, particularly if the agency has not been clear about its use of AI.

A watermark that survives editing requires us to ask whether someone actually took editorial ownership of a draft before it was published. Bylined thought leadership puts that question to its sharpest test: the named individual's expertise is the entire premise of the piece, so the responsibility for accuracy cannot be delegated to a tool.

None of this demands a new process. Agencies that already run genuine editorial oversight, checking accuracy and brand fit before publication, are already addressing much of this issue. A human editor reading for sense and context catches inconsistencies a pattern-matching checker can miss.

Don’t put too much faith in AI detectors

AI detection tools should be treated with some caution. AI checker benchmarks regularly turn up false positives, flagging work as AI-written even when no AI touched it at any stage, from brainstorming to final edit. A single tool's verdict is therefore a weak basis for an agency policy or a decision about authorship.

Detection tools exist on the other side too. Independent verification services are already testing ways to check for Anthropic's mark, and the company has said it will publish its own detection API so third parties can check content directly.

A quick Google search also shows several APIs that claim to be able to remove the watermark entirely, no matter how much human interaction has happened or not happened.

So what should PR agencies do about AI-generated content?

AI can be a useful tool for developing ideas and producing a first draft. It should not remove the need for a writer or editor to understand, challenge and take responsibility for the finished work.

A good starting point is to be clear about how much AI you used; this article, for instance, was drafted by an AI with a long and complex prompt, rewritten by one human and heavily edited by another.

Would it have a watermark if we had used the relevant version of Claude? We cannot know without testing it. But we would be comfortable explaining how the finished piece was produced and defending the editorial process, including the use of AI in its creation if required.

Just like the lemon juice writing held up to the heat, a watermark can tell you that Claude was involved somewhere in the creation of a piece of text. It doesn't tell you how involved it was, although the detection tools are only going to get more sophisticated from here.

Richard Stone is the founder of technical PR agency Stone Junction, which is a corporate affiliate member of the CIPR.

More from Richard Stone

Are your LinkedIn links costing you 60% of your reach?

The LinkedIn company page is dead. Long live the expert

How PR can turn technical Stem skills into real connection

European marketers value brand over AI visibility