Invisible Unicode Characters Explained

Unicode contains hundreds of code points that render as nothing at all. They exist for good reasons — typography, script shaping, text direction — and they cause a remarkable amount of trouble once they escape into ordinary text.

The zero-width family

Code point	Name	Purpose
U+200B	Zero Width Space	Allows a line break with no visible space
U+200C	Zero Width Non-Joiner	Prevents two characters from ligating
U+200D	Zero Width Joiner	Forces joining — builds multi-part emoji
U+2060	Word Joiner	Forbids a line break
U+FEFF	Zero Width No-Break Space	Legacy byte order mark

U+200B is the one people mean by "invisible character". U+FEFF is the classic cause of a mystery character at the start of a UTF-8 file. U+200D is load-bearing: strip it and 👨‍👩‍👧 falls apart into three separate emoji.

Invisible mathematical operators

U+2061 FUNCTION APPLICATION, U+2062 INVISIBLE TIMES, U+2063 INVISIBLE SEPARATOR and U+2064 INVISIBLE PLUS carry semantic meaning for maths software — they tell a parser that f(x) is a function call rather than multiplication. In plain prose they have no business being there at all, which makes them a favourite carrier for hidden data.

Bidirectional formatting marks

U+200E, U+200F and the range U+202A–U+202E control text direction for mixed left-to-right and right-to-left content. U+202E RIGHT-TO-LEFT OVERRIDE is notorious: it can make a file named exe.txt display as txt.exe, a trick used in real-world malware distribution. Most systems now strip or flag it in filenames for that reason.

Lookalike spaces

These are not invisible — they render as whitespace — but they are not the space character, and code comparing strings does not care how they look.

• U+00A0 Non-breaking space — the single most common intruder, injected by HTML and by Word
• U+2002–U+200A — en space, em space, thin space, hair space and the per-em fractions
• U+202F Narrow no-break space — appears around punctuation in French typography
• U+3000 Ideographic space — full-width space used in CJK text

A good cleaner converts these to a regular space rather than deleting them. Deleting an U+00A0 turns "10 kg" into "10kg" and, worse, can fuse two words permanently.

Variation selectors

U+FE00–U+FE0F select between glyph variants — most visibly U+FE0F, which turns a monochrome symbol into a colour emoji. Because there are sixteen of them and they attach silently to any character, they have become the standard vehicle for unicode steganography: an entire hidden message can be encoded in variation selectors attached to a single visible emoji, and it survives most copy-paste.

What breaks when they get in

• Exact-match search fails. The text on screen is identical; the bytes are not.
• Logins fail inexplicably. A zero-width space pasted into a password field is part of the password.
• Code will not compile. An invisible character inside an identifier produces an error on a line that looks perfect.
• URLs break. A U+200B inside a link makes it a 404 that reads correctly.
• CSV and JSON parsing corrupts. A BOM mid-file, or an U+00A0 where a delimiter was expected.
• Character limits are hit early. Invisible characters still count.

How to find and remove them

Paste the text into the invisible character detector. The Analyze tab lists every non-ASCII code point with its unicode name and count; the Visualize tab labels each one inline so you can see exactly where it sits; the Clean tab strips them, converting invisible spaces to regular spaces so word boundaries survive.

In code, the equivalent is a regex over the relevant ranges:

text.replace(/[\u200B-\u200F\u2028\u2029\u2060-\u206F\uFEFF\uFE00-\uFE0F]/g, "")
    .replace(/[\u00A0\u2000-\u200A\u202F\u205F\u3000]/g, " ")

Two passes, deliberately: delete the truly invisible, normalise the space lookalikes. Doing it in one pass is the mistake that fuses words together.

Related

Zero width space remover · Does ChatGPT watermark its text?