Invisible Characters in Text: Zero-Width Spaces, Hidden Watermarks, and How to Remove Them

A zero-width space (U+200B) is completely invisible on screen but very real in the bytes, and it, along with non-breaking spaces, the byte-order mark, and the tag characters used to watermark AI-generated text, can silently break your search, code and spreadsheets. You can’t see them, so you can’t fix them by hand, but a scanner that walks the text code point by code point reveals and removes every one. Do it privately with the invisible character detector; here’s what’s hiding and why it matters.

What’s hiding in there

The usual suspects fall into a few groups:

• Zero-width characters, the zero-width space (U+200B), joiner (U+200D), non-joiner (U+200C) and word joiner (U+2060). They occupy no visible space at all.
• Deceptive spaces, the non-breaking space (U+00A0), narrow no-break space (U+202F) and ideographic space (U+3000). They look like a normal space but aren’t.
• The byte-order mark (U+FEFF), often left at the very start of a file; it silently corrupts the first field of a CSV or the first line of code.
• Bidirectional controls (U+202A-202E), can reorder how text displays, which has been used to disguise filenames and code.
• Tag characters (U+E0000-E007F) and variation selectors, increasingly used to invisibly watermark AI-generated text by encoding a hidden fingerprint.

A quick reference table

The characters you are most likely to run into, what they look like, and what they do:

Character	Code point	Renders as	Typical problem it causes
Zero-width space	U+200B	Nothing	Breaks exact search, keyword and identifier matching
Zero-width non-joiner	U+200C	Nothing	Splits a “word” that looks whole
Zero-width joiner	U+200D	Nothing (joins glyphs)	Alters emoji sequences; confuses tokenizers
Word joiner	U+2060	Nothing	Same as ZWSP but harder to spot in old tools
Non-breaking space	U+00A0	A space	Breaks CSV parsing, trim(), number parsing
Narrow no-break space	U+202F	A thin space	Sneaks into copied numbers and dates
Ideographic space	U+3000	A wide space	Pasted from CJK sources; fails whitespace checks
Byte-order mark	U+FEFF	Nothing	Corrupts the first CSV field or first line of code
Right-to-left override	U+202E	Nothing	Reverses displayed order to disguise filenames
Tag characters	U+E0000-E007F	Nothing	Carry hidden watermark or smuggled payloads

Code points are shown in the standard Unicode U+ notation. The names come from the Unicode Standard; the “typical problem” column reflects common text-processing experience rather than any single specification.

Why they matter

These characters cause bugs that are infuriating precisely because the text looks correct:

• Search and matching fail. A zero-width space inside password means password won’t match it.
• Code breaks. A hidden character in a string literal or an identifier throws errors that don’t point to anything you can see.
• Spreadsheets misparse. A non-breaking space breaks a number into text, or splits a CSV field.
• Text gets fingerprinted. Watermark tag characters travel with copy-pasted text and can identify where it came from, a privacy consideration when you paste AI output into your own work.

A worked example

Suppose you copy a coupon code from a web page into your checkout form and it keeps getting rejected. On screen it reads SAVE20, exactly what the page advertised. But the page author wrapped the code in styling that left a zero-width space between the letters, so the real string is:

S A V E 2 0   →   S·A·V·E·2·0   (a U+200B sits after the E)

Your eyes see six characters; the server compares seven code points and finds no match. The same failure mode explains a whole family of “but it looks identical” bugs:

• A spreadsheet column of prices copied from a report imports as text, not numbers, because each value is padded with a non-breaking space (U+00A0). Sums return 0 or an error.
• A JSON file refuses to parse because a byte-order mark (U+FEFF) sits before the opening brace.
• A Python or JavaScript file throws a syntax error on a line that looks perfectly normal, because an invisible character crept into an identifier during a copy-paste.

In every case the fix is the same: reveal the code points, then strip or convert them.

Where these characters come from

Invisible characters rarely arrive on purpose. Common sources include:

• Copy-paste from rich sources, web pages, PDFs, Word documents and chat apps often carry non-breaking spaces, narrow spaces and BOMs into whatever you paste them into.
• Exports and encodings, a file saved as “UTF-8 with BOM” prepends U+FEFF; some CSV exporters emit non-breaking spaces as thousands separators.
• Deliberate obfuscation, a bidirectional override (U+202E) placed inside a filename can flip the display order of the characters after it, so an executable named to end in “exe.jpg” can appear to end in a harmless image extension.
• Watermarking and payload smuggling, tag characters and variation selectors can encode a hidden fingerprint, or even a whole hidden message, that rides invisibly inside otherwise normal text.

Why a chatbot can’t help here

This is one of the clearest cases where a tool beats an AI. A large language model operates on tokens, not raw code points. It literally cannot reliably see an invisible character, and it will happily tell you a string is “clean” when it isn’t, or invent characters that aren’t there. A scanner that iterates the text code point by code point is exact, every time. (If you want to see exactly what’s there, the Unicode character inspector shows every code point, its category and its bytes.)

How to detect and remove them

1. Paste your text into the invisible character detector.
2. It lists every hidden character it found, the code point, the name, and how many times each appears.
3. Copy the cleaned version, which removes zero-width and control characters and converts exotic spaces back to normal spaces.

For a broader cleanup of pasted or AI-generated text, straightening curly quotes, converting em-dashes, normalising whitespace and stripping hidden characters in one pass, use the text cleaner. Everything runs in your browser, which matters because the text you’re checking is often something you’d rather not upload.

Quick summary

Invisible characters, zero-width spaces, non-breaking spaces, the BOM, bidi controls, and AI-watermark tag characters, are invisible on screen but real in the bytes, and they quietly break search, code and CSV parsing and can fingerprint text. You can’t fix what you can’t see, and a chatbot can’t reliably find them, but a code-point scanner reveals and removes every one. Clean your text privately with the invisible character detector and the text cleaner.

Sources: the Unicode Standard (character categories and code points) · Unicode Technical Standard #39, Security Mechanisms · general text-processing practice. Educational information.