Remove invisible AI watermarks from text
Some AI tools and copy-paste flows embed zero-width characters, hidden Unicode tags and special spaces in text. They are invisible on screen but trivially detectable. Paste your text to see every hidden character and strip them in one click.
Cleaning happens entirely in your browser — the text never leaves your device. This removes invisible characters only; it does not change the wording or the AI score.
What is an AI text watermark?
A text watermark is a pattern hidden inside writing so that a copy can be traced back to its source. With AI-generated text the simplest version is a set of invisible Unicode characters — zero-width spaces, joiners, direction marks, tag characters, unusual spaces — sprinkled between letters and words. You cannot see them, but any script can read them, and some detectors use them as a shortcut to label text as machine-generated.
These characters also arrive by accident. Copying from a web page, a PDF, a chat interface or a word processor often drags along non-breaking spaces, soft hyphens and byte-order marks. They break search, make strings fail to match, corrupt code and spreadsheets, and can trip plagiarism and AI checkers.
This tool scans every character of your text against the Unicode ranges that are used for invisible marking, shows you exactly what it found and where, and removes it. Nothing is sent to a server — the scan and the cleanup run in your browser.
How to remove hidden characters
Instant, local and free.
Paste your text
The scan runs as you type and shows a count of hidden characters by category.
Review what was found
Toggle "Show hidden characters" to see each invisible character marked inline, so you know what is being removed.
Copy the cleaned text
Click Remove, then copy. Optionally normalize dashes and quotes at the same time.
Six kinds of invisible characters
Each category below is scanned separately so you can see how a text was marked.
Zero-width characters
U+200B–U+200D, U+2060, U+FEFF — invisible joiners and spaces inserted between letters.
Bidirectional controls
U+200E, U+200F, U+202A–U+202E, U+2066–U+2069 — direction marks that have no business in plain prose.
Variation selectors
U+FE00–U+FE0F and the supplement — modifiers that can carry hidden data.
Unicode tag characters
U+E0000–U+E007F — the invisible plane-14 characters used to smuggle data into emoji and text.
Disguised spaces
Non-breaking, hair, thin and figure spaces that look like a normal space but are not.
Soft hyphens and fillers
U+00AD, U+3164, U+FFA0 — characters that render as nothing.
Invisible to you, obvious to software
A zero-width space between two letters does not change how a word looks, but it changes the word: "report" becomes a different string from "report". Search finds nothing, a password with a hidden character never matches, a CSV column misaligns, and a tracked copy of a document is identifiable long after it has been pasted elsewhere.
The same is true for AI detection. A few checkers look for these characters first and treat them as proof of machine origin — even when the text itself is human-written and the characters came from a copy-paste. Removing them removes a false signal, not the real analysis; use the AI detector for that.
Who uses the watermark remover
Writers and editors
Clean text pulled from chat tools, PDFs and docs before publishing or submitting it.
Developers
Strip zero-width characters and odd spaces that break string matching, JSON, CSV and source code.
Students
Make sure an essay copied between apps does not carry invisible characters that confuse checkers.
Data and support teams
Fix imported records, form entries and customer messages that will not match or search correctly.
Chat windows, PDFs and word processors
Most hidden characters are not a conspiracy; they are plumbing. Chat interfaces insert zero-width joiners around emoji and formatting. PDF exports add soft hyphens where lines were broken. Word processors use non-breaking spaces after titles and figure spaces in tables. Web pages carry byte-order marks and right-to-left marks from templates.
A smaller number of tools do add characters on purpose so copies can be traced. Either way the fix is the same: scan, see what is there, and strip it before the text goes into a code editor, a form, a submission portal or a published page.
What this tool does not do
It removes invisible characters. It does not rewrite your text, and it does not change how the writing reads to a statistical AI detector — predictable phrasing stays predictable without a single hidden character. If you need the text itself to read more naturally, use the AI humanizer.
Some AI systems watermark at the level of word choice rather than characters (a statistical pattern in which words were selected). That kind of watermark is not visible in the characters and cannot be removed by a character scan. We tell you this plainly so you do not rely on the tool for something it cannot do.
Everything you need to check, fix and clean AI text
Detect AI is one of three free tools that work together. Check a text, rewrite it in a human voice, or remove the invisible characters AI tools leave behind — no account for any of them.
AI Detector
Paste any text and get an AI probability score with every suspicious sentence highlighted. Works for ChatGPT, GPT-4o, Gemini, Claude and more.
Open tool →✍️AI Humanizer
Rewrite AI-sounding text so it reads naturally. Choose a reading level, purpose and strength, then see the before-and-after AI score.
Open tool →🤖ChatGPT Detector
A detector page tuned for ChatGPT output — the tells of GPT-4o and GPT-5 writing, and how to read the result.
Open tool →Watermark remover — frequently asked questions
Is my text uploaded anywhere?
No. The scan and the cleanup run entirely in your browser with JavaScript. You can disconnect from the internet after the page loads and it still works.
Which characters are removed?
Zero-width spaces and joiners, bidirectional controls, variation selectors, Unicode tag characters, non-characters, soft hyphens and script fillers are deleted. Disguised spaces (non-breaking, thin, hair, figure) are converted to a normal space. Ideographic spaces used in Chinese, Japanese and Korean are kept.
Will it remove emoji or accented letters?
No. Emoji, accents and all visible characters stay. Only the invisible modifiers that can be attached to them are stripped; if an emoji depended on a variation selector to pick its style, it may fall back to the default rendering.
Does removing hidden characters make AI text undetectable?
No. It removes one superficial signal. Statistical detectors read the writing itself. Use the AI humanizer if the text needs to read more naturally, and be honest about AI use where disclosure is required.
Why are em dashes optional?
An em dash is a visible, legitimate punctuation mark, not a watermark — but it is also the most recognized stylistic tell of AI writing. The option lets you convert them to commas if you want; it is off by default.
Can I see what was removed?
Yes. Turn on "Show hidden characters" and each invisible character is marked inline with its code point, before you remove anything.
Does it work on code or data?
Yes — it is especially useful for source code, JSON, CSV and configuration files where a single invisible character causes failures that are very hard to find by eye.
Is there a length limit?
Because everything runs locally, there is no server limit. Very large texts (hundreds of thousands of characters) may take a second or two to scan.
Next: check how the text reads
With the hidden characters gone, paste the cleaned text into the free AI detector to see its real AI score, or into the AI humanizer to make it read more naturally.