Check any text for an AI watermark
Paste your text below and we'll use Anthropic's watermark detection API to determine if your text includes a Claude watermark. That API hasn't been released yet — results today are a preview using placeholder data. We'll ask for your email before running the check and send you a magic link — no password needed.
How this works
Conventional AI detectors inspect finished writing and guess from style — how predictable the word choices are, how uniform the sentences feel. That approach is unreliable enough that many institutions have stopped trusting it.
A watermark works from the other direction. As the model generates text, it subtly biases its word choices according to a secret key, leaving a statistical pattern that does not change the meaning and is invisible to a reader. A detector holding that key can measure whether the pattern is present. It is closer to checking for a signature than to guessing at a writing style.
Reading a result
The two outcomes are not equally informative, and this is the part people most often get wrong:
- Watermark found — strong evidence. The pattern is very hard to produce by accident.
- No watermark found — proves much less than it appears to. It is consistent with a human author, but equally consistent with a model that does not watermark, a watermarking model with the feature switched off, text that was paraphrased or edited, or a passage too short to measure.
What happens to your text
Your text is sent to the detector for that single check and then discarded — it is not written to our database or our logs. We keep only the word count, the timestamp and the verdict. It is never used for training and never sold. The about page covers this in full, and the privacy policy spells out every service involved.