Free browser tool

AI Watermark Remover

Paste any text to see the invisible characters hiding inside it. The scan names every mark it finds, decodes hidden messages when they are there, and removes only the categories you choose.

Runs in your browser Nothing is uploaded No account needed

Scan and clean

Your text stays in this tab and is never submitted.

Processed locally
0 characters
0 characters

What the scan found

Marks found
0
Marks removed
0

Scan some text to see the breakdown.

    Where the marks are

    Your text with every hidden character replaced by its code point. Hover a chip to see what it is.

    Nothing to show yet. Hidden characters appear here once the scan finds them.

    What to remove

    The scan always reports everything it finds. Removal is your choice, one category at a time.

    Smuggled payload carriers

    These have no place in ordinary prose. They are the channels used to hide data inside text.

    Invisible and format characters

    Rarely needed in ordinary writing, so they are removed by default. Check the result if your text is right-to-left.

    Depends on your text

    These characters have real uses in emoji, multilingual text, and typography. They are reported but left in place unless you turn them on.

    What a character scan can and cannot do

    Removing characters is not removing a watermark

    Any tool that claims otherwise is selling you something. Here is the honest split.

    1

    Verifiable

    Invisible characters, format controls, and smuggled payloads are countable. The scan names each one, shows you where it sits, and the counts after cleaning are yours to check.

    2

    Best effort

    Statistical watermarks live in word choice. The model biases which token it picks, so the signal is spread across the whole text and no character edit reaches it. Rewriting is the only lever, and a rewrite cannot exceed the quality of the model doing the rewriting.

    3

    Out of scope

    Content Credentials (C2PA), EXIF, and XMP live in image, PDF, and document files rather than in text. Pixel and audio watermarks are a separate layer again. Pasting text here does not touch any of them.

    Invisible does not mean harmless

    What is actually hiding in the text

    A hidden character is one that occupies a position in the text but renders as nothing. Copy that text anywhere and the character travels with it. Some arrive by accident, pasted in from a PDF or a rich-text editor. Others are placed deliberately: the Unicode tag block and variation selectors each encode a byte per character, which turns ordinary-looking text into a carrier for a hidden message. The scan reports both kinds and tells you which is which.

    What this tool does not promise

    This tool finds and removes characters. It is not an AI detector, it cannot tell you who wrote a text, and it cannot promise that any detector will change its verdict.

    Use it to clean text you own and to inspect text you were given. Do not use it to conceal authorship where disclosure is required.

    Provenance marking is arriving

    What AI companies are actually marking

    The EU AI Act's transparency rules became applicable on August 2, 2026, and the accompanying code covers machine-readable marking for generated text, images, audio, and video where it is technically feasible. Providers are responding in different ways.

    Google publishes SynthID Text, a statistical watermark applied while the model chooses its next token. Anthropic is rolling out text marking for supported Claude models and Content Credentials on some file exports. OpenAI's public provenance work currently shows up as labels and file-level credentials rather than a published text-sampling spec.

    None of that is a hidden character you can delete. That is the distinction this tool is built around: character-level marks are countable and removable, model-level marks are not.

    Characters are not provenance

    This scanner reports characters. It cannot verify authorship, read file metadata, or remove a model-level watermark, and it does not claim to.

    Read the European Commission guidance

    How the watermark scan works

    The original and the cleaned version stay side by side, so you decide what leaves.

    1

    Paste the text

    Nothing is uploaded. The scan runs in this tab, on your machine.

    2

    Read the scan

    Every hidden character is named, counted, and shown in position. If it was carrying a message, you see the message.

    3

    Choose what to remove

    Turn categories on or off, compare the two versions, then copy the one you want.

    AI Watermark Remover FAQ

    Answers about what the scan detects, what it cannot touch, and how your text is handled.

    Clean the text. Then see how AI talks about you.

    Clean the characters here, then use LLM Pulse to track how ChatGPT, Perplexity, and Gemini actually describe your brand.

    Start Free Trial
    14-day free trial
    Cancel anytime

    Not ready to buy yet? Schedule a demo and discover LLM Pulse with our founders