This tool is honest about what it can and cannot prove. Here is the full list, written without marketing language.
What it can find
- Invisible formatting characters such as zero-width spaces, joiners, soft hyphens, and byte order marks.
- Bidirectional control characters that can alter visual reading order.
- Special whitespace that looks like a normal space but has a different code point.
- Words that mix writing systems in ways that can spoof identifiers or URLs.
- Characters that would change under standard Unicode normalization.
What it cannot find
- Statistical watermarks like SynthID Text that change model output probabilities without inserting characters.
- Image pixel watermarks or C2PA metadata embedded in photos or videos.
- Whether a human or an AI wrote the text.
- Plagiarism, paraphrasing, or rewriting.
How to read the results
Each finding is a signal, not a verdict. A zero-width space might be a watermark, but it might also be an accidental copy from a PDF. A BiDi control in code is suspicious; the same control in Arabic prose is normal. Use the report to decide what to review, not what to believe.
Remember: "No hidden characters found" only means the scanner did not detect the signals it knows about. It does not prove the text is human-written or unmarked.