Text Cleaner_

Set Em/en dashes to the replacement you want — a hyphen, a comma, a plain space, or nothing at all — and every — and – in the text is converted as you type. The tool handles the spacing around them too, so “word — word” becomes “word - word” rather than “word - word”. The em dash became a talking point because language models produce it far more often than most people type it: they learned from edited prose, where it is standard punctuation.

The ones that actually turn up are zero-width spaces (U+200B), zero-width joiners and non-joiners, word joiners, byte-order marks, soft hyphens, narrow no-break spaces (U+202F) and variation selectors. Most arrive by accident: they come from web pages, PDFs, editors and tokenisation, not from a deliberate mark. A run of variation selectors or Unicode tag characters (U+E0000–U+E007F) can carry hidden data, which is why they are worth seeing rather than guessing about. This page lists every one it finds, with its code point and count, and lets you decide.

Use the AI output preset. It removes invisible and direction-control characters, converts the special spaces to ordinary ones, straightens curly quotes, turns em dashes into hyphens, strips the Markdown (**bold**, ## headings, link syntax, code fences and tables) and normalises the bullet glyphs to -. What you get is the same words in plain text that will paste anywhere without surprises.

Because it is not the characters you think it is. Word gives you curly quotes instead of straight ones, non-breaking spaces instead of spaces, en dashes instead of hyphens, and sometimes a byte-order mark at the very start of the text that makes the first column fail to match anything. The Pasted from Word preset converts all of them; the inspector shows you which ones were there in the first place.

That is two problems at once, and the Pasted from a PDF preset fixes both. Words split across lines with a hyphen are re-joined (“exam-\nple” → “example”), and the hard line breaks inside paragraphs are removed while the blank line between paragraphs is kept. Lists and headings are left on their own lines, because joining those is what makes other “remove line breaks” tools produce a wall of text.

It reads your input and lists every character that is not plain ASCII: the character itself, its code point, its Unicode name, how many there are, and one line on what it breaks. Invisible and direction-control characters are shown with a placeholder, because by definition you cannot see them. Each row has a button that switches on the option which removes that kind of character, so you can go from “something is wrong with this text” to a fix without knowing any of the terminology.

U+200B is a character with no width: it is there, it takes up a position, and it shows nothing. Web pages use it to allow a line break inside a long word; copy-paste then carries it along. The damage is quiet — a search for “invoice” will not match “invo​ice”, two identical-looking strings compare as different, and a primary key that contains one is a bug nobody can see.

They tell the renderer to display text in a different order than it is stored. That is essential for Arabic and Hebrew, and it is also the basis of the “Trojan Source” trick, where source code reads one way to a human reviewer and compiles to something else, and of file names like “gnp.exe” that display as “exe.png”. If you find one in a pasted snippet that has nothing to do with right-to-left languages, it deserves a look.

A homoglyph is a character that looks like another one. Cyrillic а, Greek ο and full-width ａ are different letters from Latin a, o and a, and no amount of staring at them will tell you which is which. They are used in lookalike domains and in text meant to defeat copy-detection, and they arrive innocently from transliteration and bad OCR. The inspector names them and the cleaner can fold them back to Latin.

U+00AD is an invisible hint that says “you may break the word here”. Justified documents and PDFs are full of them. They are invisible until the word wraps, which means a word can contain one and still look perfectly normal while failing every search and comparison you throw at it.

A non-breaking space (U+00A0) looks identical and behaves differently: text never wraps at it, split(' ') does not split on it, and trim() in some languages does not trim it. It is the most common invisible troublemaker in pasted text, along with the narrow no-break space (U+202F) that turns up in French typography and in model output.

Set Unwrap lines to paragraphs. Lines inside a paragraph are joined into one, and the blank line that separates paragraphs is kept. Set it to everything instead if you genuinely want one long line — useful for a CSV cell or a single-line prompt, and rarely what you want for prose.

There are two switches, because there are two questions. Remove duplicate lines is exact. Ignore case and spacing treats “Jane@Example.com ” and “jane@example.com” as the same line, which is what you want for a list of addresses. Keep only duplicates inverts the whole thing and shows you what repeats — the fastest way to find the rows a spreadsheet exported twice.

Use the E-mail list preset, or set Extract to e-mail addresses. Every address in the text is pulled out, duplicates dropped, one per line (or comma-separated, if that is what the next tool wants). The same extractor handles URLs, domains, numbers, IP addresses, hashtags, mentions, dates and quoted strings.

The Slug preset strips accents, lowercases, and replaces everything that is not a letter or a number with a hyphen — so “Café — Crème Brûlée!” becomes cafe-creme-brulee. Check the result before publishing: transliteration of non-Latin scripts is a judgement call no tool makes perfectly.

Yes, both ways. Join lines with puts them on one line with a comma, semicolon, pipe, tab or space between them; Split on does the reverse. Quote every line wraps each one in quotes, with a CSV option that adds the trailing comma, which is how most of this ends up being used: turning a column from a spreadsheet into something you can paste into an IN (…) clause.

No. The page is static HTML and every transformation happens in JavaScript in your tab. There is no backend, no request when you type, no analytics on the content, and nothing is stored. You can open the developer tools and watch the network stay silent, or disconnect entirely — it keeps working. That matters for the text people actually paste into cleaners: contracts, customer lists, incident reports.

Only the ones you ask it to. Everything under whitespace, characters and typography touches formatting, not content. The transforms that do change words — remove punctuation, remove numbers, case conversion, slugify, extraction — are separate switches, off by default, and the change log lists every step that fired.

Tens of thousands of lines are fine; the work is done as you type, so very large files make typing feel sluggish long before anything fails. For a multi-megabyte file, paste it in sections. There is no upload limit because there is no upload.

Because U+200D is doing a job there. A family emoji is several people glued together with joiners; a profession emoji is a person joined to an object. Strip the joiners and you get the pieces back. That is why the inspector names the character and counts it instead of quietly deleting it — in ordinary prose it is noise, inside an emoji it is the glue.

Cleaning never edits your input: the left pane stays exactly as you pasted it, and the right pane is the result. Switch an option off and the output goes back. Use as input is the one deliberate exception — it moves the cleaned text across so you can run a second pass on it.

Free, no account, no ads, no upload limit. It is one of the free browser tools at jasperbernaers.com.