AI is now watermarking your text. Here's what that actually means.

Karen From Marketing

Claude just started quietly stamping an invisible watermark into every bit of text it writes. Gemini has been doing it for a while. ChatGPT is about to join in.

If you've ever used AI to help write anything, this now applies to you. So let's cut through it.

What's actually happening

Every time Claude writes something, it's now nudging certain word choices in a pattern only a detector can spot. You can't see it. I can't see it. It doesn't change what you read. It's just sitting there in the words, invisible, waiting for a matching key to pick it up.

Why now? The EU brought in a law called the AI Act, and the big AI companies signed up to watermark their output to stay compliant. Anthropic, Google, OpenAI, Meta, Microsoft, all on board. And because building one version for Europe and one for everyone else is a headache, they're just rolling it out worldwide.

So whether you're in Brisbane or Berlin, your AI-assisted text now carries this invisible tag.

Here's the bit that matters for your business

It doesn't know the difference between “AI wrote this” and “AI proofread something I wrote myself.” Ran your own email through Claude just to tidy the grammar? Still gets the mark. That's a pretty big asterisk on what this watermark actually proves.

And it doesn't say anything about quality. A watermark can't tell you if the content is accurate, well-argued or worth reading. It just tells you AI touched it at some point. A brilliant, carefully edited piece and a lazy, unchecked one carry exactly the same invisible tag. So if you were hoping this fixes AI slop, it doesn't. It just makes bulk AI output traceable. That's a narrower job than the headlines suggest.

And here's the part nobody's talking about enough

Within about a week of this going live, people had already built tools to strip these watermarks out. Free ones. Open source ones. Whole websites dedicated to it.

Even Anthropic's own engineers admitted it, on the record: “it's not perfect, you can edit it, but it's a first step.” Translation: a decent rewrite or heavy edit will probably beat it.

So we've got a brand new transparency measure that a weekend coder can apparently work around. Make of that what you will.

What I'd actually do about it

Nothing dramatic. This isn't a reason to panic or ditch AI tools. It's a reason to keep doing what good marketers should be doing anyway: editing properly, checking your facts, and being upfront with clients about where AI sits in your process.

A watermark was never going to replace good judgement. It still won't.

If you want the full breakdown with the sources and the technical detail, my team at Oceania Marketing Group wrote it up properly. Worth a read if you want the receipts.