Anthropic has started marking the text Claude produces, so platforms can tell it was written by AI. Worth knowing before you publish under your own name.
On 14 August 2026, Anthropic confirmed it is adding an invisible text watermark to Claude's output. The mark is applied worldwide, not just in the EU, and it uses SynthID-Text, a method developed by Google DeepMind. You will not see it in the text. It sits in the wording itself, and a detector that knows how to look can read it and flag the text as coming from Claude.
The change is Anthropic's way of meeting Europe's transparency rules. The company signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026, and the EU AI Act asks providers to mark AI-generated text. Rather than run two versions of Claude, Anthropic switched marking on for everyone.
Timing matters here. Models launched on or after 2 August 2026 carry the watermark going forward. For Claude models released before that date, the law allows a transition period, and Anthropic says it is working to add watermarking to those too, rolling out over the coming months. So the coverage is widening rather than fully in place on day one.
One practical detail on how sturdy the mark is. Light editing probably will not remove it, and it survives a copy and paste. A complete rewrite, where every word is replaced, will strip it out. In plain terms, tidying a draft leaves the signature intact.
Why this matters if you publish under your own name
If your firm uses Claude to draft website copy, marketing, or client-facing material, the text you put out can now be identified as AI-assisted by any platform that checks for the mark. This is a provenance question, not a penalty. Nothing here says a search engine will rank you lower or that an AI assistant will cite you less. It simply means AI involvement in your writing is more detectable than it was.
The calm response is to treat it the way you would treat any question of authorship. Decide where AI drafting is fine and where a piece needs to be genuinely yours, then edit accordingly. For anything that carries your name and your judgement, the value was always in the thinking and the final human pass, and that is exactly what a light edit preserves and a real rewrite makes wholly your own.
Images are handled differently and more gently. For supported file types (PNG, JPG, and SVG), Claude attaches a C2PA content credential, a small signed note in the file's metadata saying Claude helped make it. Anthropic is clear that this is very different from a watermark: it carries no identifying information about you, only that Claude was involved.
Frequently asked questions
Do I need to do anything before publishing content I drafted with Claude?
No action is required. The watermark is invisible and does not change how the text reads or works. The sensible step is simply to decide which pieces you are happy to publish as AI-assisted and which you want to make fully your own with a proper edit.
Will this hurt my search rankings or how AI tools cite me?
There is nothing in Anthropic's announcement to suggest it will. The watermark is about provenance, letting platforms detect that text came from Claude. It is not a ranking signal and it does not change how you are cited.
Can I remove the watermark by editing the text?
Not with light edits. Anthropic says the mark survives copy and paste and light editing. Only a complete rewrite, where every word is replaced, removes it fully. If a piece needs to be entirely your own, rewrite it rather than tweak it.
Does this apply to me if I am outside the EU?
Yes. Although the change is driven by EU transparency rules, Anthropic applies the watermark globally, so it reaches all Claude users in Australia, the UK, and everywhere else.
What about images I create with Claude?
Images use a different mechanism. For PNG, JPG, and SVG files, Claude adds a C2PA content credential in the file's metadata. Anthropic stresses this is not a watermark and holds no identifying information, only a note that Claude helped produce the file.
