The job
Where the AI watermark detector fits in your workflow
Between copying text out of a chat window and pasting it into whatever comes next. That gap is where invisible characters travel, because they are real characters and the clipboard carries them faithfully into your CMS, your spreadsheet, your code editor and your email client.
Running the AI watermark detector at that point costs nothing and takes one paste. It reports 14 invisible characters by codepoint and count, and highlights 17 phrasing patterns that read as machine-written, all inside your browser with nothing transmitted and no cap on how often you do it.
Know the difference
Two different things go by the name AI watermark
Worth settling before you judge any tool in this category, because the word is used for two unrelated mechanisms.
- Invisible Unicode characters
- Actual characters in the text: zero-width spaces, word joiners, odd space variants. They survive copy and paste and a program can locate each one exactly. This is what the detector finds.
- Statistical watermarking
- A model biasing its choice between near-equivalent next words in a consistent hidden pattern, recoverable only by a matching statistical test. It lives in the wording, not in any character. The detector does not do this and neither does any general-purpose tool.
- Signed file metadata
- A cryptographic manifest attached to a generated file recording how it was made. That is a property of files rather than pasted text, and it is not something this tool reads.
Workflow
How to run the AI watermark detector in three steps
Unlimited, so there is no reason not to do it before anything ships.
- Step 1
- Paste the text exactly as copied. Retyping or passing it through a plain-text editor first strips the characters, which defeats the point of scanning.
- Step 2
- Work through the markers. Each invisible character shows its codepoint, its name and how many times it appears, so you can see whether you have one stray zero-width space or forty.
- Step 3
- Treat characters and phrases differently. Characters are almost always safe to strip. Phrases are prompts to edit, so rewrite the genuinely empty ones and leave the rest.
Why it matters
What these characters break once they are downstream
The symptoms rarely look like a character problem, which is why they cost hours. The failure shows up in a tool that has no idea an invisible character exists.
- Search stops matching
- A zero-width space inside a word means find-and-replace, spell-check and site search all miss it, while the word looks perfectly normal on screen.
- URLs and slugs break
- An invisible character in a slug produces a link that looks right and resolves to nothing.
- Data files misparse
- A byte-order mark in the middle of a document confuses parsers, feeds and validators, and CSV and JSON imports fail on a field that looks fine.
- Wrapping goes wrong
- A no-break space keeps two words glued together, and a soft hyphen adds a hyphen to a word that never had one as soon as the line wraps.
- Character counts inflate
- Invisible characters occupy slots, so a field with a hard limit rejects text that visibly fits.
The count
Why this AI watermark detector lists 14, not 87
Competing tools advertise 34, 87 or over 100. The larger figures are reached by counting visible characters: em dashes, curly quotation marks, variation selectors. Those are things you can see on the screen, so calling them hidden stretches the word past usefulness.
The 14 here are invisible or space-impersonating characters only. Three of them, the no-break space, the narrow no-break space and the hair space, are handled as space substitutes rather than deletions, because removing them outright would fuse two words. Em dashes are a separate matter with a separate tool, since a dash is punctuation you chose, not a marker you did not.
Before you strip
The findings you should not act on
Not every hit is contamination. These characters have real typographic jobs, and a workflow that strips everything automatically will eventually break something that was correct.
A soft hyphen in a long word may be doing deliberate hyphenation control. A no-break space may be holding a figure to its unit on purpose. A zero-width non-joiner in Persian or several Indic scripts is carrying meaning rather than noise, and that text needs reviewing rather than cleaning. The detector handles one case for you already: a zero-width joiner between two emoji is what fuses them into a single glyph, so it is checked against its neighbours and never reported.
The other half
The 17 phrasing patterns, and why they outlast the characters
Invisible characters vanish the moment anyone retypes a sentence or pastes through a plain-text step. The phrasing does not, which is why the detector flags 17 documented patterns alongside the characters, each with a label and a reason.
The set covers the scene-setting opener about a rapidly evolving landscape, the hedge that announces information rather than giving it, the Moreover and Furthermore chain, delve, the tapestry and realm and testament metaphors, the not-just-X pivot, unlocking potential, seamless and robust and cutting-edge, when it comes to, plays a crucial role, and the email opener hoping this finds you well. Each is a prompt for your judgement, not a rule.
Straight answer
Will cleaning this text get it past an AI detector?
No, and we will not suggest it. Detectors work on the writing, so invisible characters are largely irrelevant to what they report. People who build detection tools say the same thing publicly.
Scan and clean for the reasons that actually pay off: text that searches correctly, links that resolve, data that imports, fields that accept what visibly fits, and prose without the stock phrases everyone now recognises. Those are all verifiable improvements. A detector's verdict is not something any tool can promise.