Paste your draft and GPTCleanup scans it for the AI-sounding patterns and giveaways, rewrites them so your writing reads naturally, and removes the hidden characters copied ChatGPT text carries. See every finding before anything changes.
Cleaned drafts
end up here
Medium
Substack
WordPress
Ghost
Hashnode
dev.to
Quora
Reddit
WattpadPlatform names and marks indicate where drafts cleaned with GPTCleanup are commonly published. All marks are the trademarks of their respective owners. No partnership, sponsorship, or endorsement is implied.
The ChatGPT watermark detector scans pasted text for the watermark characters AI detection systems look at: zero-width spaces, zero-width joiners, and non-breaking spaces left behind by ChatGPT, Claude, Gemini, Grok, and other AI assistants.
Use it as an AI watermark remover, hidden character remover, or invisible character cleaner: one pass covers all three. Zero-width spaces, word joiners, byte-order marks, soft hyphens, variation selectors, and tag characters all come out, while your visible words stay exactly as written.
Every watermark character is reported by Unicode code point with a count, the way a proper ChatGPT watermark checker should, and each flagged phrase is quoted back with the pattern it matched. No black box: you see the full list before the ChatGPT watermark remover changes anything.
Em dashes, en dashes, two-em dashes, and three-em dashes are detected separately from hidden characters. Replace them with a regular hyphen or remove them with smart spacing, or leave them alone.
An AI text watermark is anything in the output that identifies how the text was produced. In practice that takes two forms, and only one of them is invisible.
The first is character-level. Invisible Unicode markers, zero-width characters, and non-breaking spaces get inserted by the chat interface as it formats the reply, then travel with the text when you copy it. They are the digital fingerprint people mean when they search for an AI watermark remover, and they are trivial to strip once you can see them.
The second is statistical, and it is the one AI content detectors and AI detection tools actually use. Rather than looking for hidden characters, detectors measure how predictable the word choices are: stock openers, filler transitions, hedged phrasing, and uniform sentence rhythm. That signal lives in the visible words, which is why no invisible character remover can touch it.
A ChatGPT watermark checker that only reports hidden characters answers half the question. This one runs both passes so you can see the whole picture before you decide what to change.
421 characters in total: 37 named individually below, plus 384 more across the variation selector and tag-character ranges. This is the supported set, written out, not a claim to catch every marker a model or an editor might leave behind.
These have no width at all. They survive copy and paste, they are invisible in every editor, and they are what people usually mean by a ChatGPT watermark.
Removed completely
U+200BZero-width spaceU+200CZero-width non-joinerU+200DZero-width joinerU+2060Word joinerU+FEFFByte-order markU+00ADSoft hyphenU+180EMongolian vowel separatorSpaces that are not the ordinary space character. Identical on screen, but they break find-and-replace, wrap oddly, and split words in search indexes.
Replaced with a normal spaceU+00A0Non-breaking spaceU+202FNarrow no-break spaceU+205FMedium mathematical spaceU+3000Ideographic spaceU+2000En quadU+2001Em quadU+2002En spaceU+2003Em spaceU+2004Three-per-em spaceU+2005Four-per-em spaceU+2006Six-per-em spaceU+2007Figure spaceU+2008Punctuation spaceU+2009Thin spaceU+200AHair spaceU+2028Line separatorU+2029Paragraph separatorFormatting controls that set text direction. Harmless to read, obvious to anything inspecting the underlying characters, and never intentional in ordinary prose.
Removed completelyU+200ELeft-to-right markU+200FRight-to-left markU+061CArabic letter markU+2066Left-to-right isolateU+2067Right-to-left isolateU+2068First strong isolateU+2069Pop directional isolateU+202ALeft-to-right embeddingU+202BRight-to-left embeddingU+202CPop directional formattingU+202DLeft-to-right overrideU+202ERight-to-left overrideFamilies used to smuggle invisible payloads into text. A tag-character sequence can encode an entire hidden message that renders as nothing, so these are removed wholesale rather than listed one by one.
Removed completelyU+FE00 to U+FE0FVariation selectors 16U+E0100 to U+E01EFVariation selectors supplement 240U+E0000 to U+E007FTag characters 128Not in the way the word suggests. OpenAI has never shipped a cryptographic signature in ChatGPT text output, and nobody has demonstrated one. The hidden characters people find are real, but they are residue rather than a mark: the web interface inserts them for layout and rich-text formatting, and they ride along when you copy.
That distinction matters because it tells you what cleaning them achieves. You are removing formatting debris that breaks find-and-replace, confuses screen readers, and trips form validation. You are not defeating a tracking system, because there is not one to defeat.
It also means their absence proves nothing. Plenty of ChatGPT output carries no hidden characters at all, and plenty of human writing pasted through Word or a CMS carries the same ones.
Paste the text above, and when em dashes are found a control appears offering to replace them with an ordinary hyphen or remove them with the spacing kept. Four steps in full:
Long dashes are picked up by the same pass that reads your phrasing, because an em dash is a punctuation habit rather than a hidden character. All four forms are covered: en dash, em dash, two-em dash, and three-em dash.
Each one is flagged in place and quoted back, so you can see whether the dash is holding a clause together or standing in for a comma the model could not commit to.
Take the suggested rewrite, write your own, or dismiss the finding and keep the dash. Nothing is applied until you accept it.
Words are never welded together when a dash comes out. Spacing is corrected so the sentence still reads properly.
–En dashbecomes-—Em dashbecomes-⸺Two-em dashbecomes--⸻Three-em dashbecomes---Find and replace works if you would rather not paste the draft anywhere. Search for the dash character itself and replace it with a hyphen. Watch for all four forms: the en dash is shorter and easily missed, and the two-em and three-em dashes survive any search that only looks for the standard em dash.
The catch is that find and replace cannot tell a dash holding a clause together from one standing in for a comma the model could not commit to. It flattens both, so read the result rather than trusting it.
The em dash is the most recognisable ChatGPT habit that is not hidden at all. The model reaches for it constantly, often several times a paragraph, where a comma or a full stop would read more naturally, and readers notice it long before any detector does. It is also legitimate punctuation that plenty of people use deliberately, so stripping it without asking would be editing your voice rather than cleaning your text.
This is where most tools called a ChatGPT watermark remover stop, and it is why drafts come out of them still reading as machine-written.
In today’sU+200B rapidly evolving landscape, it is important to noteU+00A0 that organisations must delve into the intricacies of digital transformation.U+2060
In today’s rapidly evolving landscape, it is important to note that organisations must delve into the intricacies of digital transformation.
Every invisible character is gone, and the four AI-sounding tells, the part a detector scores and a reader notices, are flagged for the rewrite.That is the whole problem in one line. The invisible characters came out, and the sentence still opens with an AI-sounding phrase, still hedges before it says anything, and still reaches for delve and intricacies. None of it is hidden, which is why GPTCleanup also scans the writing itself and rewrites it.
AI content detectors score exactly those word choices, and so do readers. That is why the detector here runs a second pass over the writing itself, quoting each flagged phrase back with the pattern it matched, so you can see what is actually giving the draft away. Detection costs nothing. Full rewriting is the paid part.
Copy straight out of ChatGPT and paste it in. Do not retype or reformat it first, because that can strip the very characters the ChatGPT watermark detector is looking for.
Two passes run over the draft at once. One reports every hidden character by Unicode code point and count. The other reads the writing itself and flags the phrasing patterns that survive character cleanup.
Nothing is changed behind your back. Each finding shows what it is and why it was flagged, and you choose what to keep and what to fix.
The ChatGPT watermark remover applies what you accepted. Your words, names, numbers, and structure stay exactly as they were.
It is the invisible Unicode characters that travel with text copied out of ChatGPT: zero-width spaces, zero-width joiners, non-breaking spaces, byte-order marks, and formatting attributes. They are invisible to a reader but plainly present in the underlying characters. The term is also used loosely for the phrasing fingerprint, the stock openers and filler transitions the model repeats, which is the part that survives every character cleaner.
Not a cryptographic one. OpenAI researched statistical text watermarking and has never shipped it in ChatGPT, and no third party has demonstrated a hidden signature in its output. The invisible characters people find are real but incidental: the web interface inserts them while formatting the reply, and they ride along when you copy. Their absence proves nothing, and their presence does not prove ChatGPT wrote the text.
The ChatGPT watermark detector makes two passes. The character pass compares your text against 421 supported characters, covering zero-width spaces, zero-width joiners and non-joiners, word joiners, look-alike spaces, directional controls, variation selectors, and tag characters, and reports each match by Unicode code point with a count. The phrasing pass reads the writing itself and flags stock openers, filler transitions, hedges, tier-one AI verbs, and punctuation habits including long dashes. Visible text is preserved either way.
Stripping invisible characters from your own text is ordinary formatting, no different from removing stray tabs, and it is legal. What does not change is any obligation you are already under: platform terms, disclosure rules, and academic or workplace policies on AI assistance still apply to the finished text. Cleaning the characters is a formatting step, not a way around a rule.
They are in different positions. Google watermarks Gemini text with SynthID, and Anthropic began watermarking Claude text in August 2026 using the same statistical approach. Both live in word choices rather than characters, so no cleaner reaches them. OpenAI researched the idea and never shipped it, and xAI has documented nothing for Grok, so ChatGPT and Grok output carries interface residue only. The hidden-character set is shared across all four; the phrasing habits are not, which is why each has its own page.
Scan the draft and every long dash is flagged in place: en dash, em dash, two-em dash, and three-em dash. Replace each with a plain hyphen, write your own wording, or keep it. Removal corrects the spacing so words never run together, and multiple consecutive dashes are handled as one. Nothing is changed until you accept it, because an em dash is legitimate punctuation plenty of people use on purpose.
No, and any tool promising that is guessing. AI content detectors score how predictable your word choices are, not whether invisible characters are present, so a draft stripped of every zero-width space can still read as machine-written. That is why the phrasing pass exists. What you get is a clear view of the patterns that read as AI, and control over which ones you change.
No. Word processors, CMS editors, PDF copy-paste, and translation tools all leave the same characters behind. Their presence tells you the text passed through software, not which software and not who wrote it. Treat a character scan as a formatting inspection rather than an authorship verdict.
No. Zero-width and directional characters render as nothing, so removing them cannot alter a word. Look-alike spaces become ordinary spaces, which keeps words apart rather than joining them. Phrase-level rewrites are shown in context and applied only where you accept them.
Yes. Detection is free and needs nothing but your text. Full phrase-level rewriting is the paid part of GPTCleanup, and it is a separate job from stripping characters.
Only one does. Google’s SynthID genuinely watermarks Gemini text by adjusting word probabilities during generation, and Anthropic began doing the same for Claude text in August 2026. OpenAI researched the idea and never shipped it, and xAI has documented nothing for Grok, so what turns up in ChatGPT and Grok output is interface residue rather than a deliberate mark.
The same hidden-character set applies across all of them, because these characters come from interfaces and editors rather than the models. What differs is the phrasing, and that is what each page covers in that model’s own voice: Claude watermark remover, Gemini watermark remover, and Grok watermark remover.