Comment: AI watermarking forces us to define authentic authorship

The question over what’s slop and what’s not is getting louder. It’s becoming more difficult to discern where human authorship ends and AI-generated content begins and, for many, that’s a problem.

Under Article 50 of the European Union’s AI Act, providers of generative AI systems are required to ensure that AI-generated outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Anthropic is among the providers responding to those requirements by introducing an electronic watermark into text generated by supported Claude models.

And here’s the thing: the watermark doesn’t just apply to fully-AI written copy. It applies to anything drafted, edited, or even proofed via Claude, and will survive copying, pasting and, to some degree, editing too.

That leaves agencies with an interesting problem. A piece of copy can carry evidence of AI involvement but a watermark can only indicate that Claude was likely involved with the content at some point; it cannot distinguish between Claude writing a piece and heavily editing it. 

So, a question: what does authentic authorship actually mean when it comes to modern content workflows? Where do copywriters stand when it comes to AI-assisted copy?

Authentic authorship hasn’t simply been a case of one person alone at a keyboard for a long time now. Research, tools, templates and style guides have all contributed to the shape and sound of a finished piece of copy, along with team and client feedback. Now, AI is part of that process too.

The World Federation of Advertisers research found that 78% of global brands surveyed are using AI-generated or AI-enhanced content in consumer-facing marketing, with 80% of those using it for marketing copy.

Claude’s watermark makes that contribution more visible, which puts a little more pressure on agencies to explain how the work came together. Suddenly, a part of the content creation process that previously happened behind the scenes is there to be seen. 

And perhaps that is where the conversation around watermarking becomes more useful. Once AI involvement is visible, agencies have to be much better at defining what human authorship actually contributed to the work both internally and with clients.

That means talking about where human judgement sits in the process, because this is where the real work happens: knowing which idea earns a place on the page, which line needs cutting, which client concern needs answering before it is even raised, and which tone will land with a specific audience. A watermark can flag that AI touched a piece of text but it struggles to tell you who decided that sentence was right.

That gap matters. The presence of a watermark gives us information about process, while authorship is also a question of judgement. For agencies, being able to account for that judgement is going to become part of demonstrating the value of the work.

That distinction is something agencies need to get comfortable articulating. Clients will ask questions. They may already be doing so. And there is good reason to think they will care about the answer: 82% of multinational brands surveyed by the World Federation of Advertisers said transparency around AI use is essential for protecting brand reputation, while 79% said it is key to maintaining consumer trust.

And Claude is unlikely to be alone for long. Around 190 organisations have now signed the EU Code of Practice on Transparency of AI-Generated Content, with several other major AI providers among them. As those platforms work out what transparency looks like in practice, agencies need to be ready for AI involvement to become increasingly visible. 

Agencies that are ahead of the game will have a clear and honest answer to, “How was this made?” in their locker, rather than scrambling about to construct one later down the line.

That answer needs to go further than naming the tools that were used. Agencies should be able to explain what role those tools played in the work and, crucially, where responsibility for the finished piece sat. The more visible AI becomes within the production process, the harder it will be to rely on the finished copy alone as evidence of the craft behind it.

Great content has always involved collaboration. As AI becomes another visible part of that process, agencies will need to be more open about how that collaboration works.

Claude’s watermark gives that conversation a new urgency because it creates a trace of AI involvement that sits beyond the agency’s own description of the process. 

One of the more interesting consequence of watermarking for agencies is it asks them to be much more specific about where their value sits in a world where AI is already part of the workflow. The agencies that can answer that confidently will have a much stronger story to tell about the work they produce, and the people behind it.

Ross Lowe is senior copywriter at Brave Bison