A visible watermark is easy to recognize. An invisible text watermark can instead encode a pattern through zero-width spaces, joiners, directional controls, variation selectors, or other Unicode format characters placed between ordinary letters. If different recipients receive different patterns, a later copy may contain evidence that helps distinguish one version from another.
This matters when copied text comes from sensitive reporting material, legal discovery, an internal investigation, or a restricted review draft. The useful first question is narrower than ‘who leaked this?’: does the text contain invisible code points, exactly what are they, and where do they occur? Those are deterministic facts this checker can report without sending the text anywhere.
A detected character is not proof of tracking. Editors, PDF extraction, bidirectional writing systems, accessibility tooling, and ordinary copy pipelines can introduce some of the same characters for legitimate reasons. The checker cannot establish who inserted a character, why it is present, whether different recipients received different versions, or whether the original document was monitored.
This release checks pasted text only. It does not open a PDF or DOCX, inspect metadata, OCR an image, compare document versions, or examine layout-layer objects. Copy the relevant text from the document, paste it below, and treat the result as inspectable evidence rather than attribution.
Edge cases worth knowing
- A zero-width character introduced by document conversion
- PDF extraction and editor conversion can leave format characters that were never intended as a fingerprint. The exact code point and position are real; the intent behind them cannot be inferred from the character alone.
- Text from Arabic, Hebrew, or another bidirectional script
- Directional controls may be legitimate and necessary for correct display. Their presence deserves inspection when unexpected, but it is not inherently suspicious or evidence of tracking.
- A clean pasted excerpt
- No finding means the pasted text contains none of the invisible characters covered by the detector. It says nothing about document metadata, images, fonts, layout objects, recipient-specific wording, or characters lost during copying.
Common mistakes
- Calling any invisible character a deliberate watermark without considering editing, language, and conversion context.
- Assuming a detected pattern identifies a person when no comparison set or recipient mapping is available.
- Treating a clean pasted excerpt as proof that the original PDF or DOCX contains no other tracking mechanism.
- Uploading sensitive source or discovery material to an unknown server when a local pasted-text check is sufficient.