How to remove an AI watermark in Word and Google Docs

Text pasted from ChatGPT or another AI tool can bring hidden characters into your document — zero-width spaces, no-break spaces, and Microsoft’s Windows AI watermark. Word and Google Docs can find and delete them with Find & Replace. What they can’t touch is the statistical watermark in the wording itself; for that you need a rewrite.

Facts checked against the sources below on October 31, 2026.

What Find & Replace can and can’t remove

• Can remove: hidden characters. Zero-width spaces, word joiners, byte-order marks, soft hyphens, no-break and narrow no-break spaces — and Microsoft’s Windows AI text watermark, which is made of four of these characters.
• Can’t remove: statistical watermarks. ChatGPT’s textGrain, Google’s SynthID and Claude’s watermark add no characters at all. They are in the choice of words, so the only fix is to change the words.

Microsoft Word

1. Open Replace: Home → Replace, or Ctrl+H. On a Mac, Edit → Find → Replace. Click More and make sure Use wildcards is unchecked — the codes below only work without it.
2. Type a code in “Find what” from the table, put the replacement in “Replace with” (leave it empty to delete), and click Replace All. Repeat for each code.
3. Microsoft’s Windows AI watermark is the four characters in a row: find ^u8203^u847^u8288^u8203 and replace with nothing. The individual codes catch it too.

Word’s ^u code takes the character’s number in decimal, so U+200B (hex 200B) is ^u8203. You can’t type or paste a zero-width character into the Find box reliably, which is why the code is the dependable route. Leave zero-width joiners (U+200D) alone: they hold multi-part emoji together.

Google Docs

1. Open Find and replace: Edit → Find and replace, or Ctrl+H (⌘+Shift+H on a Mac), and tick Match using regular expressions.
2. Delete the invisible characters. In “Find”, enter the pattern below, leave “Replace with” empty, and click Replace all.

   [\x{200B}\x{200C}\x{2060}\x{FEFF}\x{034F}\x{00AD}]

3. Turn no-break spaces into normal spaces. Find the second pattern and type a single space in “Replace with”.

   [\x{00A0}\x{202F}]

Google Docs uses RE2 regular expressions, where \x{200B} matches a character by its hex code point. The “Replace with” box takes plain text only — in Docs, regex syntax there is inserted literally — so leave it empty or type the space itself.

Removing the statistical watermark

Once the characters are gone, the watermark that the big AI providers actually use is still in the wording. It survives light edits; OpenAI’s own test found it detectable in 66% of passages even after one word in ten had been swapped for a synonym. Rewriting is what removes it:

1. Open your document in Unmark. Use “Open .docx” in the tool — the file is read in your browser — or paste the text in.
2. Click Remove watermark. Each paragraph and list item is rewritten on its own, so headings, bullets and numbering come back where they were, and every hidden character is stripped in the same pass.
3. Paste the result back and re-apply any bold, links or styles, since the result is plain text. Re-read it to make sure it still says what you mean.

Sending it as a PDF

Clean the text before you export: a PDF keeps whatever characters were in it. The PDF also adds something of its own. Word’s Save as PDF copies the document properties into the file, including the author’s name, unless you untick Document properties under Options. To check or strip that from a finished PDF, PaperKit’s Remove PDF Metadata shows the author, title, creating app and XMP data and removes them in your browser, without uploading the file.

FAQ

Sources

• Word MVP Site — Using the zero-width space (^u8203) in Word
• Google Docs Editors Help — Find and replace, including regular expressions (RE2)
• OpenAI Help Center — textGrain adds no hidden characters
• Microsoft Learn — Watermarking in Windows AI text APIs