Guides
Claude Watermark Vs ChatGPT Watermark
Claude watermark vs ChatGPT watermark is currently a comparison between something shipped and something shelved.
Anthropic began marking all Claude output on 2 August 2026. OpenAI has been reported for years to hold a highly accurate text detector it has chosen not to release.
Anthropic marks everything, globally. OpenAI has reportedly built but not released comparable tooling.
Runs in your browser. No signup, no upload, nothing stored.
Where each vendor stands
Anthropic: marking live since 2 August 2026 across Claude, the API, Claude Code, Claude Cowork and Claude Tag, applied worldwide, no opt-out, detector not yet public.
OpenAI: reporting has described an internal text watermarking system with very high stated accuracy, held back from release over concerns about the effect on users and on the product's competitiveness.
Why Anthropic moved first
The forcing function is EU AI Act Article 50, which applies to any provider serving EU users.
Anthropic chose global application rather than geofencing, which turned a regional obligation into a product-wide default and removed the option of quietly complying only where required.
The commercial tension
Marking is a straightforward competitive disadvantage in a market where a segment of users specifically wants undetectable output.
Coverage after the announcement recorded exactly that reaction, and some of it was loud. A vendor that marks loses those users to one that does not, which is why nobody did it voluntarily first.
What to expect next
The same obligation lands on every major provider serving the EU, so assume convergence rather than divergence over the next year.
Be careful about paraphrasing Claude output through another assistant on the assumption that it launders provenance. You may simply be swapping one provider's mark for another's, with an extra quality loss for free.
The practical difference today
For text you have already produced, the vendor matters. Claude output from a supported model after 2 August 2026 carries a mark; equivalent output from other assistants may not yet.
That gap is temporary and is not a strategy. Building a workflow around which vendor has not shipped marking yet means rebuilding it when they do.
Why detection status is the sharper comparison
Both companies are reported to hold detection capability, and neither has released it publicly. That is the more interesting symmetry.
Marking without detection changes nothing observable today. It changes everything on the day a detector ships — and because marking is already running, that switch applies retroactively to everything published in the meantime.
Source
The claims on this page are drawn from primary documentation and reporting rather than from other tools’ marketing copy.
TechCrunch: Anthropic says it will watermark text generated by its AI models →Related
Check what your text and files actually carry
Runs in your browser. No signup, no upload, nothing stored.