CleanPaste.site is a real tool that does roughly what its name says: you paste text, it takes out invisible and non-standard characters, and you copy the result back. It is not the only one, and the differences between these tools are smaller and more specific than their marketing suggests. What actually separates them is which codepoints each one reaches, whether you need an account, and what happens to your text after you press the button.
This page compares CleanPaste.site with the free cleaner on this site and with zerowidthspace.me, which people often find in the same search but which does something completely different. Everything below was checked on the live sites in September 2026. Where something could not be checked from outside, it says so.
What is CleanPaste.site and what does it actually do?
Key takeaways
- CleanPaste.site is a working invisible-character cleaner. Its free tier asks you to create an account and sign in before it will clean anything, which is the main practical difference from the tool on this site.
- The cleaner on this site is free with no account and no sign-in step, and it also folds lookalike letters and styled letterforms back to plain ASCII, which CleanPaste's own description of what it strips does not mention.
- CleanPaste has features this site does not: saved cleanup history, CSV export, a plagiarism and similarity check, and a paid tier that rewrites your text with a language model.
zerowidthspace.me is not an alternative. It hands you a single zero-width space to copy. That makes it the fastest way to build a test string for any of these tools.- You can settle which tool reaches further in about ten seconds by counting characters before and after. Instructions are below.
CleanPaste.site is a single-page web app. Its own page description says it removes hidden metadata and AI watermarks from generated text. The interface gives you a text box, a "Reveal Hidden Characters" view that highlights what it found and names each codepoint, and a clean button.
Its documentation page on invisible characters states what the cleaner takes out: all zero-width spaces from U+200B to U+200F, byte order marks (U+FEFF), and all format control characters in the Unicode Cf category. That list covers the characters that cause most of the real damage. The revealer built into the page reaches wider: the highlighting code in the shipped JavaScript matches C0 and C1 control characters, U+00A0, the whole U+2000 space block, bidirectional overrides, U+202F, variation selectors and the private use area.
One structural point worth knowing: CleanPaste does the cleaning on its server, not in your browser tab. Pressing clean sends your text to a hosted function. Its privacy policy says text you submit remains your intellectual property, but it does mean the exact set the server strips cannot be read from outside. Their documentation page is the best available statement of it.
Is CleanPaste.site free?
Partly, and the shape of it matters more than the word. There is a free tier for basic character cleaning, and paid tiers above it that raise the limits and add the extras described below. Check their pricing page for the current figures, since prices move and we would rather send you to the source than quote a number that has changed.
So basic character cleaning costs nothing in money. It costs an account: press clean while signed out and you get a sign-in prompt. If you clean text once a month, that is a fair trade for saved history. If you clean text nine times an hour while editing, it is friction.
What the cleaner on this site does differently
Two differences, and neither is a value judgment about CleanPaste.
The first is the front door. The cleaner at aitextwatermarkremoval.com has no account, no sign-in and no usage counter on the web tool. Paste, clean, copy, leave. Nothing about your text is stored.
The second is the character table. Invisible codepoints are only half of what AI output carries. The other half is characters that are perfectly visible but wrong: a Cyrillic U+0430 sitting where a Latin a belongs, a Greek U+03BF standing in for o, fullwidth forms, and the mathematical alphanumeric block starting at U+1D400 that gets used as fake bold in places that do not support real formatting. Those survive a Cf-category sweep intact, because they are ordinary letters as far as Unicode is concerned. They are also the ones that break exact-match search and make two identical-looking strings compare as different. This cleaner folds them to their plain equivalents, and offers em dashes and curly quotes as an optional extra pass.
Where CleanPaste is ahead: it remembers what you cleaned. If you want a record of every document you ran through a tool, exportable to CSV, this site has nothing like that and is not trying to. CleanPaste also bundles a plagiarism and similarity check, which is a genuinely separate product that most character cleaners do not offer at all.
Side by side: CleanPaste, this cleaner, and zerowidthspace.me
| | CleanPaste.site | aitextwatermarkremoval.com | zerowidthspace.me |
|---|
| What it is for | Removing invisible characters, plus paid extras | Removing invisible characters, odd spaces, lookalikes and styled letters | Copying one zero-width space to your clipboard |
| Price for basic cleaning | Free | Free | Free |
| Account needed | Yes | No | No |
| Stated character coverage | Zero-widths U+200B to U+200F, U+FEFF, Unicode Cf category | Zero-widths, exotic spaces, bidi and tag characters, homoglyphs, fullwidth and math alphanumerics | Not applicable, it produces U+200B rather than removing it |
| Lookalike letters | Not mentioned in its description | Yes | No |
| Where the text goes | Their server | Their server, processed in memory, not stored | Nowhere, the page is static |
| History and export | Yes, more on paid plans | No | No |
| Paid tiers | Paid tiers above the free one | Web tool free, API from $19 a month | None, optional tip links |
The honest summary of that table is that the first three rows matter and the rest is preference. Both cleaners will take a zero-width space out of your paragraph. The question is what else is in your paragraph.
How do you test any paste cleaner in ten seconds?
Do not take anyone's feature list on trust, including this one. Character cleaning is one of the rare things you can verify yourself in seconds.
Build a test string. The fastest source for a real zero-width space is zerowidthspace.me, which exists purely to put one on your clipboard. Paste it between two letters so you get a, then the invisible character, then b. That string is three characters long and looks two characters long.
Now run it through whichever cleaner you are testing and count the result. Any character counter will do. If the output is two characters, the tool reached the codepoint. If it is still three, it did not.
For a harder test, open your browser console and build a string that covers several categories at once:
U+200B zero-width space, the baseline caseU+FEFF byte order mark, which travels well through copy and pasteU+00A0 non-breaking space and U+202F narrow no-break space, which look like ordinary spaces and are notU+2063 invisible separator, a Cf character outside the zero-width rangeU+0430 Cyrillic small a, which is where most cleaners stopU+1D400 mathematical bold capital A, the fake-bold letterform
Paste the cleaned output back into the console and run [...s].map(c => c.codePointAt(0).toString(16)) over it. You will see exactly which of those six survived. That is the whole comparison, done properly, in one line, with no marketing involved.
What most people get wrong about cleaning pasted text
"Just paste it into Notepad first." This is the single most repeated piece of advice on the subject and it does not do what people think. A plain text editor strips rich formatting: bold, fonts, colours, links. It does not strip Unicode codepoints, because those are the text. A zero-width space that goes into Notepad comes out of Notepad. So does a Cyrillic U+0430 dressed up as an a. The round trip removes the styling and leaves every invisible character exactly where it was, which is worse than doing nothing, because now you think the problem is solved.
Here is the contrarian part, and it cuts against the tools rather than for them. More aggressive stripping is not better stripping. The intuitive ranking says the cleaner that deletes the most codepoints wins. It does not. Several of these characters are load-bearing.
U+200D, the zero-width joiner, is what holds multi-person emoji together. Delete every U+200D from a paragraph and a single family emoji shatters into three separate people. The same character is structural in Arabic, Persian and several Indic scripts, where removing it changes how letters connect. U+00AD, the soft hyphen, is a legitimate typesetting instruction. Right-to-left and left-to-right marks are not decoration in Hebrew or Arabic text; they are what stops mixed-direction sentences rendering backwards.
If your text is English prose from a chat window, a wide sweep is fine and nothing will break. If your text is multilingual, contains emoji you care about, or is going into a system where exact bytes matter, the right cleaner is the one that is specific about what it touches, not the one with the longest list. Check the output, not the feature count.
One more correction, because it decides which tool you actually need. Rewriting text and cleaning text are different operations. CleanPaste's paid tiers include a feature that paraphrases your text with a language model, which changes your word choice. Character cleaning does not change a single visible word. If you want your sentences rewritten, that is a writing tool. If you want the same sentences with the junk codepoints gone, that is a cleaner. Buying one expecting the other is the most common mismatch in this category.
Which one should you use?
Use CleanPaste.site if you want a log of everything you have cleaned, want the plagiarism check in the same place, and do not mind creating an account. Its revealer view is genuinely nice for seeing what is in a document before you change it.
Use the cleaner here if you want to paste and go without signing in, and if your text may contain lookalike letters or styled letterforms as well as invisible ones. If you are cleaning at volume from a CMS or a script rather than by hand, the same cleaner is available over HTTP.
Use zerowidthspace.me when you need to produce a zero-width space rather than remove one, which is exactly what you need to test either of the others.
Try it against your own text
The free cleaner at aitextwatermarkremoval.com takes invisible Unicode, non-standard spaces, lookalike letters and styled letterforms out of pasted text. No account, no sign-in, nothing stored. Paste the test string from the section above and count the characters that come back.
Frequently asked questions
Is there a CleanPaste.site alternative that does not need an account?
Yes. The cleaner on this site runs without any sign-up and does the same core job of stripping invisible and non-standard characters, plus lookalike letters and styled letterforms. The trade is that it keeps no history of what you cleaned, which is one of the things CleanPaste's account gives you.
Does cleaning characters change what an AI detector says about my text?
No, and any tool that tells you otherwise is describing something it cannot do. Detectors such as GPTZero, Turnitin, Originality.ai and Copyleaks score the statistical properties of word choice. Invisible characters are not part of that calculation. Removing U+200B from a sentence leaves every word exactly where it was, so there is nothing for a word-choice model to notice.
Do these tools remove a statistical watermark?
No. Statistical watermarks are real and they are a different mechanism entirely. Google has watermarked Gemini output with SynthID since 2024, and Anthropic announced SynthID-Text for Claude on 11 August 2026. Both live in which words the model picks, not in hidden codepoints, so a character cleaner has nothing to grab hold of. OpenAI has never shipped a text watermark at all.
Why does AI text contain these characters if it is not a watermark?
Mostly because of what the model was trained on and what happens between the model and your clipboard. Web pages are full of U+00A0 and typographic spacing, so models reproduce it. Chat interfaces render markdown to HTML, and copying from HTML can pick up characters that were never in the raw output. In April 2025 people noticed o3 and o4-mini emitting the narrow no-break space U+202F where an ordinary space belonged; OpenAI described it as a quirk of large-scale reinforcement learning, and it stopped within days.
Is zerowidthspace.me a paste cleaner?
No. It is a static page whose only function is to put one zero-width space on your clipboard. It removes nothing. It shows up in searches for clean paste tools because of the subject matter, and it is genuinely useful here, just for building a test string rather than for cleaning one.