AI Text Cleaner for ChatGPT, Claude and Gemini
Paste anything you copied from an AI chat and get plain, clean text back. Hidden characters, Markdown symbols, curly quotes and em dashes are removed, and your wording is never changed. Nothing is uploaded: it all runs in your browser.
Clean up
AI text cleaner examples
Three things that come along when you copy from an AI chat. The right-hand side of each is the tool's real output with its default settings.
Markdown symbols in a pasted answer
AI chats write their answers in Markdown so they display with headings and bold text. Paste one into a plain-text field, an email app that ignores formatting or a CMS, and you get the symbols underneath: asterisks around bold words, hash marks before headings, star bullets. The cleaner removes the marks and keeps the structure, so the heading becomes a plain line and list items keep a simple dash.
Copied from the chat
### Follow-up on our call Hi **Sarah**, Thanks for your time today. Quick recap: * **Budget:** approved for Q1 * **Next step:** send the draft by *Friday*
After the cleaner
Follow-up on our call Hi Sarah, Thanks for your time today. Quick recap: - Budget: approved for Q1 - Next step: send the draft by Friday
Hidden characters you cannot see
Zero-width spaces, byte-order marks and narrow no-break spaces take up no visible room, so the text looks normal. They are still characters: they break search and replace, word counts, form validation and code, and they travel with every copy. The tool lists each one it finds by name and code point, then removes it or turns it into an ordinary space.
Copied from the chat
The report is readyU+200B and attached.U+202FLet me knowU+200B if you have any questions.U+FEFF
After the cleaner
The report is ready and attached. Let me know if you have any questions.
Curly quotes, ellipses and em dashes
Curly quotes and the one-character ellipsis are correct in print, but they turn into odd symbols in some fonts, code and older systems. Em dashes have become the punctuation many readers associate with AI writing. You choose what the dashes become: commas, hyphens, or nothing changes.
Copied from the chat
“It’s not about speed—it’s about trust,” she said…
After the cleaner
"It's not about speed, it's about trust," she said...
Looking to erase text from a picture instead? That is a different job: remove text from a photo.
What is an AI text cleaner?
An AI text cleaner is a tool that removes the formatting left behind when you copy text out of ChatGPT, Claude, Gemini or another AI chat: hidden Unicode characters, Markdown symbols, curly quotes, em dashes and stray spacing. It changes how the text is encoded and punctuated, not what it says. Paste in, copy out, and the text behaves like something you typed yourself.
Most of this comes from how chat apps display answers. They write in Markdown so the answer shows up on screen with headings and bold text, and anywhere that does not read Markdown gets the symbols instead. Typographic characters such as curly quotes, ellipses and no-break spaces are correct in print but get mangled by plain-text fields, code and some fonts.
This cleaner runs entirely in your browser. The text you paste is never uploaded, stored or sent to an AI model, so it is safe for drafts you would rather keep private.
Last updated:
What the cleaner removes, character by character
Each character, where it usually comes from, and what happens to it. Anything not listed is left exactly as it is.
| Character | Code point | Usually comes from | What the cleaner does |
|---|---|---|---|
| Zero-width space | U+200B | Text copied from web pages, chat apps and word processors | Removed |
| Zero-width no-break space (BOM) | U+FEFF | Files saved with a byte-order mark, some editors | Removed |
| Zero-width joiner and non-joiner | U+200D, U+200C | Emoji sequences, Persian and Indic scripts, stray copies | Removed, except inside emoji and in scripts that need them |
| Word joiner, soft hyphen | U+2060, U+00AD | Word processors and automatic hyphenation | Removed |
| Narrow no-break space | U+202F | Number formatting, French typography, generated text | Replaced with a normal space |
| No-break, thin and other spaces | U+00A0, U+2009 and others | Word processors and web pages | Replaced with a normal space |
| Tag characters | U+E0000 to U+E007F | Hidden text and prompt-injection tricks | Removed, except in the England, Scotland and Wales flags |
| Direction marks | U+200E, U+200F and others | Text that mixes left-to-right and right-to-left | Removed unless the text contains Arabic, Hebrew or another right-to-left script |
| Curly quotes and primes | U+2018, U+2019, U+201C, U+201D | AI chats, Word, Google Docs | Straight ' and " |
| Ellipsis | U+2026 | Autocorrect and AI chats | Three periods |
| Em dash | U+2014 | AI chats and autocorrect | A comma, a hyphen, or kept: your choice |
| En dash | U+2013 | Number ranges and autocorrect | A hyphen in ranges like 10-20; treated as an em dash when it has spaces around it |
| Markdown marks | ** # ` > --- | AI chat answers | Removed; list items keep a plain dash |
How to clean text copied from an AI chat
- 1
Paste the text
Paste what you copied from ChatGPT, Claude, Gemini, Copilot or any other AI chat into the left box, or click Paste. Up to 200,000 characters at a time.
- 2
Choose what to clean
Everything except emoji is on by default. Pick what em dashes become: commas, hyphens, or leave them as they are.
- 3
Copy the clean text
The right box updates as you type. Read the summary of what changed, then copy the result and paste it wherever it is going.
Does ChatGPT hide a watermark in its text?
Not in the way the rumour says. Every so often a post claims that ChatGPT hides invisible characters in its answers so teachers and detectors can spot them. Characters like narrow no-break spaces and zero-width spaces do sometimes appear in AI output, just as they appear in text copied from web pages and word processors. They are a side effect of formatting, not a reliable signature: plenty of AI text has none, and plenty of human text does.
AI detectors do not look for them either. They score how predictable the wording and sentence structure are. Removing hidden characters, quotes and dashes changes none of that, so a detector gives the same result before and after cleaning.
Some text watermarks do exist, and they are not characters at all. Google's SynthID text watermark, which Google applies to Gemini output, works by nudging which words the model picks, so the signal lives in the wording itself. No formatting cleaner can remove it, this one included.
So use this tool to fix how text looks and behaves when you paste it. If you are asked to disclose AI use, disclose it.
When not to use it
Five cases where cleaning is the wrong move, or where you should change a setting first.
Not a rewriter or humanizer
It never changes your wording. If you want sentences rewritten, this is the wrong tool, and cleaning will not change what an AI detector reports.
Destinations that read Markdown
GitHub, Reddit, Discord and many note apps turn Markdown into formatting. If that is where the text is going, switch off Markdown symbols before you copy.
Source code
Code can contain asterisks, typographic characters inside strings and deliberate non-breaking spaces. Paste prose here. Code blocks inside an answer are kept as they are, but whole files are not what it is for.
Typeset print
Curly quotes, real ellipses and em dashes are correct in books, magazines and designed pages. For those, switch off quotes and set dashes to Keep.
Scripts that rely on joiners
Persian, Arabic and Indic scripts use zero-width joiners and non-joiners to shape letters. The cleaner keeps them there, but check the result if a text mixes scripts inside single words.
Frequently asked questions
What is an AI text cleaner?
An AI text cleaner is a tool that removes the formatting left in text copied from ChatGPT, Claude, Gemini and other AI chats: hidden Unicode characters, Markdown symbols, curly quotes, em dashes and stray spacing. It changes how the text is encoded and punctuated, not what it says, so the result pastes like text you typed yourself.
Is this AI text cleaner free?
Yes. There is no sign-up and no daily limit. Each paste can be up to 200,000 characters, which is roughly 30,000 words, and you can clean as many times as you like.
Is my text uploaded or stored?
No. The cleaning runs in your browser with JavaScript, and the text you paste is never sent to our servers or to an AI model. Once the page has loaded you can go offline and it keeps working, which is an easy way to check.
Does it remove the ChatGPT watermark?
It removes hidden characters, which is what most "ChatGPT watermark" posts are about. Those characters are a side effect of formatting rather than a reliable signature, and AI detectors do not rely on them. Real text watermarks, such as Google's SynthID, sit in the choice of words, and no formatting tool can remove them.
Will the cleaned text pass AI detectors?
No, and it is not built to. Detectors score how predictable the wording and sentence structure are. Cleaning changes characters and punctuation, not wording, so a detector gives the same result before and after. If you are asked to disclose AI use, disclose it.
Why does text from ChatGPT have asterisks and hash marks in it?
AI chats write answers in Markdown, where **two asterisks** mean bold and # marks a heading. The chat window turns that into formatting on screen. Copy the answer into a place that does not read Markdown and you see the raw symbols. The cleaner removes them and keeps the line breaks and list structure.
Why turn em dashes into commas?
Most of the time a comma reads naturally where a dash was, and it is the option people ask for when they want text to read less like AI output. When a dash introduces a list or an explanation, a colon may read better, so look over those sentences. Choose Hyphen or Keep if you prefer.
How can I see the hidden characters in my text?
Paste it in. Under the boxes the tool lists every hidden character it found by name and code point, such as Zero-width space U+200B. Click Show where they are to see each one marked in place in your original text.
Clean your next paste
Paste the text, check what changed, copy it out. It takes a few seconds and nothing leaves your browser.
Related tools
AI Image Humanizer
The same idea for pictures: make an AI image look like a real photo.
Remove AI Look
Take the glossy, over-graded finish off an AI-generated image.
Remove Text from Photo
Erase captions, date stamps and signs from an image.
AI Image Detector
Check whether an image was likely made with AI.
All of these are part of the sparkpix.ai AI photo editor, so one account works across every tool.
Last updated: 2026-10-30