The job
What a Perplexity watermark detector is for
A Perplexity answer rarely goes straight from the screen to a reader. It goes into a brief, a slide, a ticket, a CMS draft or a spreadsheet, and it is that second hop where invisible characters cause trouble. The Perplexity watermark detector sits between the two: paste the answer, see everything the visible text is hiding, then move it on clean.
The scan reports fourteen invisible Unicode characters by codepoint and marks each one where it sits. It also highlights the stock connectors and hedges that read as machine-written, so a draft that needs an edit pass is obvious before anyone else reads it. Everything happens in your browser, which means a confidential research answer is safe to paste.
Before you start
Two different things are called perplexity
Worth clearing up, because it decides whether this tool is the one you want. Perplexity is an AI answer engine. Perplexity is also a statistical measure of how predictable a passage is to a language model, and some AI detectors feed that number into a score.
This page handles the first. It reads text you copied out of the Perplexity answer engine and reports what is hidden in it. It returns no score and no percentage, because the engine behind this page does not calculate one. If you need a number, nothing here will produce one honestly.
The task
How to check a Perplexity answer for hidden characters in three steps
The fastest place to run this is the moment before you paste into the destination, not after the destination has already mangled it.
- Step 1. Copy the whole answer
- Take the citation markers with it. Numbered references and the spacing around them are a common place for invisible characters to sit.
- Step 2. Paste and read the chips
- Each finding appears inline as a chip naming its codepoint, so you can see whether the problem is one stray character or spread through the whole answer.
- Step 3. Clean, then paste onward
- One click removes the invisibles and turns the space look-alikes into normal spaces. Copy the cleaned version into your document.
The list
The hidden characters this Perplexity scan reports
Fourteen characters, grouped the way the engine treats them. Anything outside this list is not checked, and the page does not pretend otherwise.
- Zero-width family
- U+200B zero-width space, U+200C non-joiner, U+200D joiner, U+2060 word joiner. Invisible, but they occupy slots and break matching.
- U+FEFF byte-order mark
- Harmless at the very start of a file, disruptive anywhere else. Breaks feeds, parsers and validators.
- U+00AD soft hyphen
- Shows nothing until the line wraps, then a hyphen appears inside a word.
- Space look-alikes
- U+00A0 no-break space, U+202F narrow no-break space, U+200A hair space. Cleaning replaces each with an ordinary space rather than deleting it.
- Directional controls
- U+202A through U+202E. Five characters that force or reverse text direction and quietly scramble mixed-script lines.
The cause
Why Perplexity answers carry more of this than a plain chat reply
Perplexity answers by fetching pages, quoting them and summarising them. Those pages were typeset by humans and by content systems, and both leave invisible characters behind as a matter of routine: a narrow no-break space before a unit, a soft hyphen so a long term wraps neatly, a byte-order mark added by whatever exported the block.
Material carried through that pipeline and then rendered as styled text in your browser puts those characters into your clipboard along with the words. A model answering from memory with no retrieval step has fewer opportunities to pick them up.
The second variable is the model picker. Perplexity lets you choose which model writes the answer, so two answers from the same account are not guaranteed to behave the same way. If your results are inconsistent between sessions, that is a likely reason.
The cost
What hidden characters in Perplexity text break downstream
The damage always shows up one step after the paste, which is why it is hard to trace. A zero-width space in a heading splits the keyword, and the slug your CMS generates no longer matches the title on the page. A no-break space inside a spreadsheet formula or a terminal command throws an error that points at the wrong thing.
Elsewhere the same characters make find-and-replace miss an obvious match, make a CSV import reject a column that looks perfectly normal, make a screen reader pause mid-word, and make a git diff report a changed line where nothing visible changed. None of these read as a Unicode problem until you scan.
Straight answer
Does a finding prove Perplexity wrote the text?
No. A hidden character means the text passed through something that renders rich text. That could be Perplexity, or a CMS, a PDF export, a word processor or a page somebody copied by hand. The scan reports what is in the string and nothing more.
The inverse matters just as much in practice. A clean scan does not show a human wrote it. Retyping, pasting through a plain-text editor, or simply using a model that leaves no residue all produce a clean result. Use the scan to fix a document, not to settle an argument about a person.