Zero Width Space Remover — Remove Zero Width Characters from Any Text

What zero width characters actually are

Six code points do most of the damage. U+200B (zero width space) is the most common — it sits between characters and takes up no visual space whatsoever. U+200C (zero width non-joiner) and U+200D (zero width joiner) control how adjacent characters connect, which matters for ligatures in Arabic and for emoji sequences in English. U+FEFF was originally the byte order mark, a file-encoding hint that belongs at the start of a stream; encountered mid-paragraph, it is simply invisible garbage. U+2060 (word joiner) behaves like a zero width space but also tells the line-wrapping engine to treat the two surrounding characters as inseparable. U+00AD (soft hyphen) is visible only at a line break — everywhere else it hides.

Each has a legitimate use somewhere. None of them belongs in plain prose you copied out of a chat window or a CMS.

Where they come from

ChatGPT and similar tools stream answers through a renderer that assembles fragments on the fly. The renderer uses zero width joiners to hold emoji sequences together and word joiners to keep certain phrases from breaking across lines. When you press the copy button, the browser copies the character stream, not the picture you were looking at — joiners and all.

Word processors are the second major source. Microsoft Word inserts soft hyphens automatically when it hyphenates a word; paste that paragraph into a web form and the soft hyphens travel with it. Some CMS platforms inject U+FEFF at the start of a pasted block as a side effect of their rich-text editors. Web pages with bidirectional text routinely embed directional marks that survive a copy-paste into a plain editor.

The result is text that looks right on screen but behaves strangely in every system that reads it character by character.

Why zero width characters cause real problems

Search matching is the most common failure. A database query for the exact string "ChatGPT" returns nothing if there is a U+200B between the "t" and the "G". The characters look identical in every interface; the comparison is not.

Form validation breaks the same way. An email field that passes a regex check but contains a soft hyphen will fail the mail server's own validation. A username that looks unique is not unique if one copy has a zero width space the other lacks.

Code is even less forgiving. A Python string literal with a U+200B inside it compiles without complaint and then raises a KeyError at runtime because the dictionary key was stored without it. JavaScript template literals, SQL identifiers and YAML keys all behave the same way.

Word and Excel have their own quirks. Word counts soft hyphens as characters, so a document that looks like 500 words might report 512. Excel treats a cell starting with U+FEFF as text even when it contains a number, breaking every formula that references it. Sorting a column that mixes cells with and without zero width spaces produces an order that looks random.

For SEO the consequence is quieter but persistent. A title tag split by a U+200B is two tokens to a search index, not one. The keyword you optimised for may not match what the crawler read.

How to spot them before you clean

Three clues work without any tool. First, move the cursor through the text one keystroke at a time — if it jumps two positions for what looked like one gap, there is an invisible character between them. Second, click once just before a word, hold Shift and press the right arrow; if the selection seems to grab more than one character before the first visible letter appears, something is there. Third, try Find and Replace with the exact string you can see on screen — if it finds nothing, the stored string contains characters the display is hiding.

The cleaner on this page makes the count explicit. It reports each character type found, its Unicode code point, and how many times it appeared. That report is worth reading before you copy the cleaned text, because it tells you which step of your workflow is introducing the noise.

Three things people get wrong about zero width characters

They are not a virus or malware. A zero width space cannot execute code, escalate permissions or replicate itself. It is a data character, the same category as a comma or a letter. Cleaning it out removes a formatting artefact, not a threat.

They do not spread on their own. A document that contains one will not infect documents that share a folder with it. The character travels only when you copy text that contains it.

Removing them does not damage your formatting. Real spaces, paragraph breaks, tabs and punctuation are untouched. The cleaner on this page removes only the characters in its published table and collapses runs of whitespace it was already going to collapse. The words, sentences and layout come out exactly as you wrote them — minus the invisible layer.

The only case where cleaning changes something visible is a soft hyphen sitting exactly at a line break. Remove the soft hyphen and the word may no longer break there. That is a correct outcome: the line-break point was inserted by a word processor, not by you, and your layout engine is better placed to decide where to break.

How this tool removes zero width characters

Paste your text and press Clean. The engine reads the input one Unicode code point at a time — not one byte, not one code unit, but one actual character. Every code point is checked against a published table. U+200B, U+200C, U+200D, U+FEFF, U+2060 and U+00AD are removed. U+00A0 (non-breaking space), U+2007 (figure space) and U+202F (narrow non-breaking space) are normalised to an ordinary space rather than deleted, because deleting them would silently join two words.

The report underneath the output box lists every change, grouped by character type. If nothing is found, the report says so — it does not invent rows to look busy. You can check the report against the input before you copy the cleaned text, and you can use the detector page on this site if you want the report alone, without producing a cleaned copy.

Internal links: for removing AI-generated text artefacts broadly, see the <a href="/chatgpt-watermark-remover">ChatGPT watermark remover</a> or <a href="/clean-ai-text">clean AI text</a> pages. For a general-purpose pass over any kind of pasted text, the <a href="/ai-text-cleaner-online">AI text cleaner online</a> covers the same character set.