Skip to content
AI Watermark Remove

Check Claude text for hidden characters

Paste text from Claude and see a list of every hidden character in it. No percentages and no guessing, just what is actually in the file.

0 characters
More options
Checking is free: 5 texts and 2 images a day, no signup.

Checking your text

    Does Claude watermark text?

    Anthropic has not published a text watermark for Claude, and there is no confirmed deliberate character-level mark in its text output. As with the other assistants, the honest position is that absence of a published watermark is not the same as a guarantee, and the only way to know what is in your particular document is to look at it.

    What does show up in practice is the ordinary residue of generated and copied text: occasional zero-width characters, non-breaking spaces, the typographic style markers common to all current models.

    Files produced through tooling are a different matter. Documents, images, artifacts exported from an assistant session are worth checking separately, because export pipelines routinely attach generator metadata even when the text layer is clean.

    Why check at all if there is no known watermark?

    Because "no known watermark" is a statement about the model, and the thing you are submitting is a file that has been through several other systems.

    Text that has passed through a web interface, a clipboard, a Google Doc, a markdown renderer, a word processor accumulates characters at every step. The origin of a zero-width space in your final document is very often the pipeline, not the model. It is still in your document.

    The check takes a few seconds and gives you a definite answer instead of a guess, which is a better basis for deciding whether to care.

    What the report tells you

    Every flagged character is listed with its Unicode codepoint, its official name, how many times it occurs, and a plain-English note on what it is and why it might be there. Below that, your original text is shown with each invisible character rendered as a visible chip so you can see exactly where they sit.

    The tool also separates confidence levels honestly. Zero-width and tag characters are marked as probable carriers, because they have no legitimate reason to be in ordinary prose. Unusual spaces are marked informational, because they are far more often a formatting artefact than a deliberate mark, and treating them as evidence of anything would be overreaching.

    Questions

    Does Anthropic watermark Claude output?
    There is no published or confirmed character-level text watermark in Claude output. This tool reports what is actually in your document rather than making claims about any vendor's intentions.
    What about files exported from an assistant session?
    Worth checking separately. Export and rendering pipelines frequently attach generator metadata to images and documents even when the text itself is clean. Drop the file into the image tab to see what is attached.
    Is my text sent anywhere else?
    No. It is processed in memory to produce the report and discarded when the response is sent. Nothing is written to disk, logged, or used for training.

    Related tools