Inspection tool
Claude Watermark Checker
Check pasted text for invisible characters. This finds hidden Unicode — and explains clearly why that is not the same thing as a statistical text watermark.
🔒 Processed locally in your browser. Your text is never sent anywhere.
This does not remove statistical text watermarks.
A statistical watermark is encoded in a model’s word choices, not in invisible characters. Removing hidden Unicode has no effect on it, and no browser-side tool can confirm whether one is present. NoWatermark does not claim to detect or remove Anthropic’s statistical watermark.
Start with the honest answer
If you are here because you read that AI text carries a hidden watermark, the important thing to know is that there are two unrelated ideas with similar names.
The first is invisible Unicode: real characters such as zero-width spaces that render as nothing but survive copy and paste. This tool finds and removes them, and they genuinely do appear in text pasted from all sorts of sources.
The second is a statistical watermark, which biases a model's word choices in a pattern a matching detector can recognise later. It leaves no special characters at all. Nothing running in your browser can detect it, and this tool does not claim to.
Why invisible characters matter anyway
Zero-width characters and Unicode tag characters can carry a payload inside otherwise ordinary text, and bidirectional overrides can make text display in a different order from how it is stored — a trick used to disguise filenames and links.
They also break search, matching and diffing in confusing ways. Finding them is useful whatever their origin.
Emoji are safe
Zero-width joiners and variation selectors do real work inside emoji: they are what turn separate figures into a single family emoji, and what make a heart render in colour. A naive cleaner breaks those. This one recognises them and leaves them alone.
Frequently asked questions
Does Claude watermark its text?
Anthropic has not published a consumer-facing statistical text watermark for Claude. Regardless, a statistical watermark would be invisible to a browser tool like this one — so this checker reports only what it can genuinely measure: hidden characters.
Can you detect AI-written text?
No, and this tool does not attempt it. Statistical AI-text detectors are unreliable and frequently misclassify human writing. We would rather report nothing than report a guess.
Is my text sent anywhere?
No. It stays in the page. There is no API call, and no analytics event carries your text or anything derived from it.