Invisible Character Detector — Find Hidden Unicode & Zero-Width Characters
Paste any text below and instantly spot zero-width spaces, bidirectional overrides, non-breaking spaces, and other hidden Unicode marks.
Detect Invisible Characters
Drop a plain-text file here to check and clean, or click to browse
.txt, .md, .csv, .json, .html, .css, .js, .ts, .py, .java, .go, .xml, .log
What Are Invisible Characters?
Not every character in a text file is meant to be seen. Unicode contains dozens of code points whose whole purpose is to control layout, join words, or separate scripts without leaving a visible trace. An invisible character detector finds those marks so you can decide whether they belong in your text.
The most common hidden characters fall into four groups:
- Zero-width spaces and joiners —
U+200Bzero-width space,U+200Czero-width non-joiner,U+200Dzero-width joiner, andU+2060word joiner. These can sit between visible letters without changing the visual line, but they change the underlying byte sequence. - Zero-width directional controls —
U+200E,U+200F,U+202A–U+202E, andU+2066–U+2069. These flip or isolate reading order and are the building blocks of Trojan Source attacks. - Special whitespace — non-breaking space
U+00A0, narrow no-break spaceU+202F, medium mathematical spaceU+205F, ideographic spaceU+3000, and the legacy em/en spaces. They look like ordinary spaces but break parsers and URLs. - Byte order mark (BOM) —
U+FEFFat the start of a file can signal UTF-16 or UTF-8 encoding to editors, yet it is invisible to readers and can break command-line tools.
Because these characters are valid Unicode, most editors and browsers simply render them as nothing. That is exactly why they are useful for watermarking, formatting, and sometimes for abuse.
Why Do They End Up in Your Text?
Copy-Paste from PDFs and Web Pages
PDF renderers and web layout engines insert zero-width joiners, non-breaking spaces, and soft hyphens to control line breaks. When you copy the text, those marks come along.
AI-Generated Output
Some language models emit zero-width spaces or special spaces in training-data artifacts or as part of an embedded watermark. Not every model does this, but the marks can survive into your final copy.
Localization and Mixed Scripts
Bi-directional text, right-to-left scripts, and CJK layouts need control characters. A harmless layout marker in one context becomes an invisible landmine in another.
Malicious Injection
Attackers can hide bidirectional overrides inside source code, usernames, or filenames so that the visible order differs from the logical order. That is why code reviewers need an invisible character checker.
Signs Your Text Has Hidden Characters
You cannot reliably spot invisible characters by eye. The symptoms show up elsewhere:
| Symptom | Likely Cause |
|---|---|
| A URL or identifier looks correct but returns a 404 or auth error | Non-breaking or special space in the string |
| Search returns no match for text that visually matches | Zero-width space or different normalization form |
| Code compiles differently from what the reviewer saw | Bi-directional override or homoglyph |
| A CSV or JSON parser fails on "clean" data | BOM at file start or stray separator characters |
| Two cells look identical but sort differently | Ideographic space, non-breaking space, or combining marks |
How to Detect Them
Using this invisible character detector takes three steps:
- Paste or upload the text you want to inspect. The scan runs entirely in your browser; nothing is sent to a server.
- Run the scan. The tool highlights every hidden character by category, shows the exact Unicode code point, and explains why it was flagged.
- Review the report. Switch between the highlighted view, the findings list, and the section view to see exactly where each mark sits.
You can also enable or disable rule groups. For example, if you only care about zero-width spaces, turn off the other groups and rescan.
How to Remove Invisible Characters
Detection is only half the job. After the scan, the tool shows a Cleaned Text tab that strips every flagged character in one pass. Before you copy it, you can compare the cleaned output to the original and decide whether a specific rule should stay enabled.
Three things to keep in mind:
- Preserve meaning. Removing all whitespace indiscriminately can collapse formatting. Use the rule toggles to keep intentional spacing.
- Check code carefully. BiDi controls in source code are almost always malicious or accidental; removing them is usually safe, but always re-review the line.
- Save a backup. The original text stays in the text area until you clear it, so you can re-scan with different rules if needed.
FAQ
What is an invisible character in Unicode?
It is a code point that does not produce a visible glyph by itself. Examples include zero-width spaces, joiners, bidirectional controls, and non-breaking spaces. They are legitimate parts of Unicode but can cause problems when they appear unexpectedly.
Can invisible characters carry an AI watermark?
Yes, if the watermarking method inserts zero-width characters or special spaces into the generated text. This tool finds those marks. It cannot detect statistical watermarks like SynthID Text, which alter token probabilities rather than adding characters.
Is this tool the same as an AI watermark detector?
It overlaps with one meaning of "AI watermark detector" — the kind that looks for hidden characters in text. The broader term also covers statistical and style-based detection, which this tool does not do.
Does the tool upload my text?
No. The scan and cleanup run locally in your browser. Your text never leaves your device.
Which characters does it flag?
By default it flags invisible formatting marks, bidirectional controls, special whitespace, and line/paragraph separators. You can toggle each group on or off.
Can I use this for code review?
Yes. The high-severity bidirectional control category is specifically designed to catch Trojan Source vectors before they reach a repository.
Which invisible characters can this detector find?
It reports zero-width spaces and joiners (U+200B, U+200C, U+200D, U+2060), the BOM (U+FEFF), bidirectional controls (U+200E, U+200F, U+202A–U+202E, U+2066–U+2069), non-standard spaces, soft hyphens, and line or paragraph separators. These are character-level traces, so the scan finds marks that exist as code points in the text. Statistical watermarks that leave no characters are outside what any text scan can see.
Why does copy-pasted text contain hidden characters?
Because the source was never plain text. PDF renderers, web layouts, and rich editors insert zero-width joiners, non-breaking spaces, and soft hyphens to control line breaks, and the clipboard carries those marks along with the visible letters. Some AI outputs also emit zero-width spaces, so the characters survive into your document.
Can this detector tell me whether my text was written by AI?
No. It reports whether character-level traces are present, not who wrote the text, because the same zero-width or special-space marks can come from copy-paste, PDF extraction, or malicious injection. Statistical watermarks such as SynthID Text change token probabilities instead of adding characters, so a browser text scan cannot detect them; the honest result is whether hidden characters were found, not an authorship verdict.
How do I read the scan report, and can I also clean the text?
Start with the Findings Report: each card shows the code point, severity, and surrounding context, and Critical and High findings are the ones to review first because they mark characters that can change how code, URLs, or logins are read. The Highlighted Text view marks each hit in place, and the Cleaned Text tab removes every enabled finding in one pass. Detection and cleaning stay in your browser — nothing is uploaded, stored, or logged, and there is no scan limit — but the scan only sees characters, so it cannot expose image metadata or statistical watermarks.
Related Guides
- What Is an AI Watermark, Really? Why "AI watermark detector" can mean three different things, and why character scanning is only one of them.
- Trojan Source: When Text Order Lies How bidirectional Unicode characters can hide malicious code, and what the scanner flags.
- Unicode Normalization in Plain English Why the same letter can have multiple code points and why that breaks comparisons.
- What a Character Scanner Cannot Prove The honest limits: statistical watermarks, image metadata, and the human-vs-AI question.
Related Tools
- AI Watermark Detector Full scanner for hidden Unicode, BiDi, and mixed-script characters.
- Text Cleaner Remove messy and invisible characters from any text.
- AI Watermark Detector Free The free tier: no account, unlimited scans, and everything runs locally in your browser.
- Online AI Watermark Detector Pure browser-based detection, no install, and it works right away on phones and tablets.