Perplexity watermark detector
What the Perplexity watermark detector does
Paste an answer copied out of Perplexity and the Perplexity watermark detector shows you what is in the text that you cannot see. It checks for fourteen invisible Unicode characters, marks each one where it sits, and names the exact codepoint it found. It also highlights the stock phrases that make writing read as machine-written, with a short note on why each one was flagged.
The whole check runs in your browser. The text you paste is never sent anywhere, there is no account, and there is no cap on how many answers you run through it. That matters more here than on most tools, because the thing people usually paste into a Perplexity watermark detector is a research answer they have not published yet.
Read this first
Perplexity the product, perplexity the number
The word does two jobs, and search results for it are a mess because of that. Perplexity is an AI answer engine. Perplexity is also a statistics term for how surprised a language model is by a piece of text, and several AI detectors use that number as an input.
This page is about the first one. The Perplexity watermark detector reads text copied out of the Perplexity answer engine and reports the invisible characters it carries. It does not compute a perplexity score, and it does not return a percentage. If a number is what you came for, this is the wrong tool and it is better to say so than to show you a figure the engine never calculated.
The characters
The hidden characters in Perplexity answers
These are the fourteen the detector reports. Nothing else is claimed, because nothing else is checked.
- U+200B zero-width space
- Invisible, but it still takes a character slot and survives copy and paste.
- U+200C and U+200D joiners
- Break find-and-replace, spell-check and URL slugs. Joiners inside emoji are left alone.
- U+2060 word joiner
- Usually wedged against a dash. No visual effect, but search and diffing stop matching.
- U+FEFF byte-order mark
- Mid-document, it confuses parsers, feeds and validators.
- U+00AD soft hyphen
- Invisible until a line wraps, then a hyphen appears in a word that never had one.
- U+00A0, U+202F, U+200A
- No-break, narrow no-break and hair spaces. Identical on screen, different in the bytes. Cleaning swaps them for normal spaces.
- U+202A to U+202E
- Five directional controls. Stray copies quietly reorder mixed-direction text.
How it works
How to check Perplexity text for a watermark in three steps
No setup, no upload, no sign-in. The scan is instant on the length of answer Perplexity actually produces.
- Step 1. Copy the answer
- Select the Perplexity answer and copy it. Include the citation markers if they came with it, because that is often where the characters sit.
- Step 2. Paste and scan
- Drop it in the box. Each hidden character appears as a chip naming its codepoint, and flagged phrasing is highlighted in place.
- Step 3. Clean or keep
- One click strips the hidden characters and swaps the space look-alikes for normal spaces. Your wording is not touched.
Where they come from
Why Perplexity answers pick up characters from the web
A Perplexity answer is not written from nothing. It is assembled from pages the engine fetched, quoted and summarised, and those pages are full of typographic invisibles that nobody notices while reading. A news site sets a narrow no-break space before a percent sign. A documentation page uses soft hyphens so long identifiers wrap. A publisher's CMS drops a byte-order mark at the top of every block it exports.
When that material is carried into an answer and then rendered in your browser as styled text, the invisibles ride along into the clipboard. That is the reason a Perplexity answer tends to carry more of this than a plain chat reply from a model with no retrieval step.
There is a second source, which is the model itself. Perplexity lets you pick which model answers, so two answers from the same account can behave differently. That is worth knowing before you conclude a clean scan means anything about the product.
Why it matters
What hidden characters in Perplexity text actually break
Most people find these as a bug, not as a privacy worry. A zero-width space inside a heading splits the keyword, so the URL slug your CMS generates stops matching the title a reader sees. A no-break space pasted into a shell command or a spreadsheet formula produces an error that reads like nonsense.
The same characters cause find-and-replace in Word to miss obvious matches, screen readers to pause where no pause exists, JSON and CSV imports to fail on a column that looks correct, and git diffs to show a changed line where nothing visible changed.
Straight answer
Does a finding prove Perplexity wrote the text?
No. A zero-width space means a zero-width space is present. Text that passed through any rich-text surface can carry one: a CMS, a PDF export, a word processor, a chat app, or a web page somebody copied by hand. The detector reports what is in the string, and that is the whole of what it knows.
The reverse trips people up more often. A clean scan does not prove a human wrote the text. Plenty of AI output carries no invisible characters at all, particularly if it was retyped, pasted through a plain-text editor, or produced by a model that leaves no residue. Treat the result as one signal about one file, never as a verdict on a person.
In practice
Who uses the Perplexity watermark detector, and when
Researchers and analysts who use Perplexity to gather sources and then paste findings into a report, where a stray byte-order mark will break the export. Writers checking a draft before it goes into a CMS. Developers who asked Perplexity a question about an API and are about to paste the answer into a config file.
Editors also use it in the other direction, on submitted work, to see what a document carries before deciding whether to ask about it. The useful version of that conversation starts from what the scan found, not from an accusation.