Invisible Text Watermark
Embed a hidden zero-width watermark in your text, scan any text for invisible characters, or strip them out completely. Free and private — everything runs in your browser.
Note: Embed and decode use this tool's own zero-width scheme. The Scan tab detects invisible characters from any source (ChatGPT, Claude, Gemini…), and decoders for third-party watermark formats will be added as vendors publish them. Statistical watermarks — Anthropic's Claude watermark and Google's SynthID — live in word choice, not characters, so no character-level tool can detect or remove them.
What Is an Invisible Text Watermark?
An invisible text watermark hides machine-readable data inside ordinary text using zero-width Unicode characters — symbols that take up no space and have no visible shape. The text looks exactly the same to a reader, but anyone with a scanner can detect the mark and decode the message it carries: an author ID, a document version, a tracking tag.
The idea has become mainstream with generative AI. Starting August 2026, the EU AI Act requires AI-generated content to be marked in a machine-readable way, and AI labs have shipped schemes that tag model output.
Those lab schemes are a different technique from this tool's. Anthropic's Claude watermark and Google's SynthID are statistical: they add nothing to the text and use no hidden characters, steering instead which word the model picks when several fit equally well. Only the lab holding the key can verify them, and no character-level scanner or cleaner — including this one — can see or strip them. Character-based watermarks like the zero-width scheme below are the visible-to-tools kind: you can embed them, detect them and remove them yourself.
Why Clean Text Pasted From AI Tools?
- Hidden markers travel with copy-paste — text copied from chatbots and AI editors can carry zero-width characters you never asked for, which then end up in your documents, emails and CMS.
- They can break things — invisible characters trip up search, string matching, CSV imports, code and SEO tools that treat them as real content.
- Privacy — a hidden tag can identify which account or session produced a text. Removing it before you share the text keeps that metadata to yourself.
Scan first to see what's hiding in the text, then use Remove to strip it with one click. The visible content is never modified.
How It Works
- Embed — your payload is turned into UTF-8 bytes, each byte into 8 bits, and each bit into a zero-width character (ZWSP = 0, ZWNJ = 1). The stream is wrapped in invisible delimiters and interleaved between the words of your text.
- Scan — the scanner looks for the delimiters, decodes any bit stream inside them, and lists every invisible character found, even ones from other tools.
- Remove — every zero-width and invisible character is stripped, leaving only the visible text.
Everything runs locally in your browser — no text is ever uploaded.
Specialized Watermark Tools
Guides and dedicated tools for each use case and AI assistant:
- Remove Watermark from Text — strip every invisible character in one click
- Scan Text for Watermarks — detect hidden marks before you share
- Embed Watermark in Text — tag your own writing with an invisible ID
- ChatGPT Watermark Remover
- Claude Watermark Remover
- Gemini Watermark Remover
- Grok Watermark Remover
- Copilot Watermark Remover
- Perplexity Watermark Remover
- DeepSeek Watermark Remover
- Mistral Watermark Remover
FAQ
What is an invisible text watermark?
An invisible text watermark is hidden information embedded in a piece of text using zero-width Unicode characters. Readers see completely normal text, but the characters are still there in the data, so a scanner can detect the mark and read the message it carries.
How does this tool hide a message inside text?
Your payload is converted to UTF-8 bytes, each byte becomes 8 bits, and every bit is written as a zero-width character — Zero-Width Space for 0 and Zero-Width Non-Joiner for 1. The stream is wrapped between two invisible delimiter characters and interleaved between the words of your text, so it survives copy-paste unchanged.
Can I check if AI-generated text has a hidden watermark?
Partly. Paste the text into the Scan tab and the scanner reports every invisible character it finds — the kind that copy-paste and some tools leave behind. What it cannot see are statistical watermarks like Anthropic’s Claude watermark or Google’s SynthID: those add no characters at all, they steer which word the model picks, and only the lab holding the key can verify them.
Does removing the watermark change my visible text?
No. The Remove tab only strips invisible and zero-width characters (watermark markers, joiners, BOMs, soft hyphens and similar). Every visible letter, space and punctuation mark stays exactly as it was.
Is my text uploaded to a server?
No. Embedding, scanning and removal all run locally in your browser with plain JavaScript. Your text never leaves your device.
Why would I watermark my own text?
Writers, researchers and companies embed invisible IDs in documents to prove authorship, trace leaks (each copy gets a unique tag), or verify that a text is the original. Anyone who pastes the text elsewhere carries the hidden mark with it.