How to Check If Text Has an AI Watermark (Free, in Your Browser)

"AI watermark" can mean two very different things, and you check for them in different ways. One kind is hidden characters: invisible symbols sitting between your words. The other is a statistical pattern in the wording itself, like Google's SynthID. The first you can find in seconds. The second can't be seen at all.

Kind 1: hidden characters

Unicode has dozens of characters that take up no visible space or look identical to normal ones. They can end up in text copied from an AI tool, a web page or a PDF. The common ones are:

• Zero-width space (U+200B), zero-width joiner / non-joiner (U+200D, U+200C) and word joiner (U+2060): completely invisible.
• Byte order mark (U+FEFF): invisible, sometimes left at the start of pasted text.
• Narrow and regular no-break spaces (U+202F, U+00A0): look like normal spaces. In 2025 users reported narrow no-break spaces appearing in output from some newer ChatGPT models; OpenAI described it as a quirk rather than a watermark, but the characters are real either way.
• Unicode tag characters (U+E0000–U+E007F): invisible, and able to spell out hidden messages.
• Look-alike letters (homoglyphs): a Cyrillic "а" or "о" swapped in for the Latin one. It looks the same but is a different character.

Why care? Some of these can be used to fingerprint text, some break search and spell-check, and a few plagiarism and AI-checking tools flag them as suspicious.

How to check, step by step

1. Open the free hidden character checker. It runs entirely in your browser; nothing is uploaded.
2. Paste your text.
3. Every hidden or look-alike character is highlighted, with a count of each kind. Hover over one to see its Unicode code.
4. Click Remove hidden characters, then Copy text for a clean version.

No tool handy? In Microsoft Word, turn on Show/Hide ¶ (Ctrl+Shift+8) to reveal some unusual spaces, though not zero-width ones. Developers can search for the regex [​-‍⁠  ].

Kind 2: watermarks in the wording

Some AI providers can watermark text without adding any characters. Google DeepMind's SynthID Text, used for Gemini, slightly nudges which words the model picks. Each choice still reads naturally, but across a few hundred words the pattern is statistically detectable by someone holding the key.

You can't see this kind, and a character checker can't find it, because the text is made of perfectly normal characters. The only person who can test for it reliably is the provider who holds the key.

How to remove each kind

• Hidden characters: the free checker removes them instantly.
• Wording patterns: the words themselves have to change. Google notes that SynthID's detection weakens when text is thoroughly rewritten. A full rewrite of every sentence, which is what GhostVerb's watermark remover does, replaces the word choices the pattern lives in, while keeping your meaning and voice.

A quick word on AI detectors

AI detectors (the "98% AI" kind) aren't watermark checkers. They guess from writing style, and they're known to be inconsistent, including flagging human writing. Clean characters and natural, varied wording help, but no honest tool can promise a detector result.