See what's hiding in your text. Clean it without uploading anything.
A free tool that runs in your browser. It inspects text for invisible characters — zero-width, bidi controls, Unicode tag characters — and reads metadata layers in images, PDFs, and Word files. Every finding is listed with its codepoint and position, so you can decide what to remove.
Try it now — paste text with hidden characters, or use this sample:
What exactly does it scan?
A character-by-character technical scan: zero-width characters (U+200B range), bidi controls (U+202A–U+202E), Unicode tag characters (U+E0000–E007F), variation selectors, and non-standard spaces. Files (PNG, JPEG, PDF, DOCX) are inspected at the chunk level: EXIF, XMP, text tags, and C2PA records.
What it doesn't do — plainly
This is a technical scan, not a statistical one. It cannot read a model-level statistical watermark (that requires the vendor's private key) and it never claims to detect AI writing from style. A clean result says nothing about who wrote the text — it only means no technical traces were found.
Numbers you can quote
- 128codepoints in the Unicode tag block U+E0000–E007F, each checked character by character
- 10bidi controls (U+202A–U+202E, U+2066–U+2069) that reorder visible text while staying invisible themselves
- 0bytes of your text leave your device — text cleaning is fully in-browser; file scrubbing runs in memory and is discarded after the response
Questions people actually ask
Does Unstamped detect whether AI wrote a text?
No. It detects technical traces: invisible characters, bidi controls, tag characters, and file metadata. Style-based detection is a statistical problem and is not part of this tool.
Can an AI text watermark be removed completely?
The technical layer (hidden characters, file metadata) is removed deterministically. A model-level statistical watermark cannot even be read without the vendor's key; bypassing it means a rewrite — an effort the tool explicitly labels as best-effort, never guaranteed.
What is U+200B and why does it matter?
The zero-width space: invisible when rendered, but present in the data. It is used in machine-generated text patterns (watermark injection) and in data hiding. A high count in ordinary text is a strong signal of automated processing.
Do you store my files or text?
No. Everything — text, images, PDFs, DOCX — runs in your browser's memory. Nothing is uploaded to any server; the cleaned result downloads straight back to your device. No storage, no content logs.
What is C2PA in my image?
Content Credentials: a standard that embeds a record of how a file was created or edited, and with which tools. The scan reports its presence, and the cleaner removes it deterministically from the file.
Start here
Scan text or a file
Paste text or upload a file and see every finding with codepoint and position.
Scan →Clean what we find
Deterministic removal of hidden characters and file metadata, with a download.
Clean →AI writing detector
An instant estimate of how machine-like your text reads — every raw signal shown, clearly labeled an estimate.
AI Check →