テキストから見えないAI透かしを除去

What are invisible text watermarks?

Invisible text watermarks are hidden markers embedded in digital text to identify its source or track its distribution. Unlike visible watermarks (like a logo overlaid on an image), these are completely invisible to the human eye. They work by inserting zero-width characters (Unicode characters that take up no visual space), varying the types of spaces used (regular spaces vs. non-breaking spaces vs. other Unicode space variants), or encoding information in the pattern of line breaks and punctuation. AI text generators, plagiarism detection systems, and some publishing platforms use these techniques to mark their output. When you copy text from these sources, the invisible watermarks come along silently.

How AI text watermarking works

AI language models like ChatGPT can embed statistical watermarks in their output by subtly biasing word choices in a detectable pattern. For example, the model might prefer certain synonyms over others in a way that's statistically measurable but invisible to readers. Other systems insert zero-width Unicode characters (U+200B, U+200C, U+200D, U+FEFF) at specific positions to encode a binary identifier. Some use homoglyph substitution — replacing Latin letters with visually identical characters from other alphabets (like Cyrillic 'а' instead of Latin 'a'). These techniques allow the source to later analyze suspicious text and determine with high confidence whether it came from their system. Our tool detects all of these patterns and removes them, leaving clean, natural text.

Why remove invisible watermarks?

There are legitimate reasons to remove invisible watermarks from text. If you're editing AI-assisted content into your own original work, you want the final text to be clean and not carry hidden markers. Researchers analyzing text need clean samples without watermark interference. Developers building on AI outputs need to ensure the text won't trigger false positives in detection systems. Privacy-conscious users don't want hidden identifiers tracking their content across the internet. Additionally, zero-width characters can cause problems: they break text-to-speech systems, interfere with search functionality, cause issues in code, and create inconsistencies when text is processed by different systems. Clean text is more portable, more accessible, and more professional.

Types of invisible characters we remove

Our tool targets the full spectrum of invisible text markers. Zero-width characters: U+200B (zero-width space), U+200C (zero-width non-joiner), U+200D (zero-width joiner), and U+FEFF (byte order mark/zero-width no-break space). Unusual whitespace: U+00A0 (non-breaking space), U+1680, U+2000-U+200A (various width spaces), U+202F, U+205F, and U+3000 (ideographic space). Control characters: U+0000-U+001F and U+007F-U+009F (invisible control codes). We also detect homoglyph substitutions and normalize them to standard ASCII equivalents. The tool shows you exactly what was found and removed, with a character count, so you have full transparency into the cleaning process.

Best practices for clean text

After cleaning text with our tool, always review the output to ensure the meaning and formatting are preserved — while our tool is careful to only remove invisible characters, it's good practice to verify. If you're working with AI-generated content, remember that removing watermarks doesn't change the fact that AI assisted in creation; be honest about AI use where disclosure is required (academic work, journalism, etc.). For the cleanest results, paste text directly from the source rather than through intermediate applications that might add their own formatting. If you need to preserve specific Unicode characters for legitimate reasons (like non-English text), check our settings to customize which character ranges are cleaned. Clean text isn't just about removing watermarks — it's about ensuring your content is accessible, portable, and professional.