Scan Text for Watermarks
Paste any text and reveal everything hiding inside it: zero-width watermark markers, invisible characters and their exact Unicode codepoints — with a verdict on whether a decodable watermark is present.
What the Scanner Detects
The scanner inspects your text at the Unicode level and reports two things separately:
- A watermark verdict — whether the text contains a decodable zero-width watermark (invisible delimiters wrapping a valid bit stream), and if so, the hidden payload it carries.
- Every invisible character present — a full table of hidden characters with name, codepoint and count, even when they form no watermark at all.
That second part matters: most hidden characters in real-world text are stray artifacts from rich-text copy-paste or AI output, not deliberate watermarks — but you want to know about them either way.
How Zero-Width Watermarking Works
Character-level text watermarks hide binary data in plain sight: a zero-width space (U+200B) stands for a 0 bit, a zero-width non-joiner (U+200C) for a 1 bit. Eight bits make a byte, bytes spell out a UTF-8 message, and invisible delimiter characters (U+2062 and U+2063) mark where the hidden stream begins and ends. The stream is interleaved between ordinary words, so the text looks — and reads — completely normal.
The scanner reverses exactly that process: find the delimiters, read the bits, decode the message.
The Limits of Detection
Honest answer: not every AI watermark is made of characters. Statistical schemes such as Google’s SynthID Text bias the model’s word choices instead of inserting anything — there is no hidden character to find, and no scanner can detect or remove them. What this tool guarantees is complete visibility into the character layer: if something invisible is physically in your text, the scan will list it.
Found something? The watermark remover strips every invisible character in one click.
Frequently Asked Questions
How does the watermark scanner work?
It checks the text character by character against a set of invisible Unicode characters — zero-width spaces, joiners, delimiters, BOMs, soft hyphens and more — and reports each one with its codepoint and count. If the text contains this site’s watermark encoding (invisible delimiters around a zero-width bit stream), it also decodes the hidden payload for you.
Can it detect watermarks from ChatGPT or Claude?
It detects the invisible characters users report in AI chatbot output, which is the visible evidence of character-level marking. It cannot detect statistical watermarks like Google’s SynthID, which alter word-choice patterns rather than inserting characters — no character scanner can.
What does "delimiters found but payload invalid" mean?
The text contains the invisible start and end markers of a watermark, but the data between them is corrupted or uses a different encoding — for example because someone edited the text by hand and deleted part of the hidden stream.
Is the scanned text sent to a server?
No. Scanning runs entirely in your browser. Nothing you paste is uploaded, logged or stored.