Observable text artifacts
Watermark Slayer inspects literal Unicode format controls and registry-defined structures in the exact text you provide. A finding reports an observable artifact, not who wrote the text or which system produced it.
Where the public registry marks an action supported, you can choose a reviewable cleanup edit while the original remains available. Language-sensitive and provenance-like findings stay report-only.
Statistical text watermarking
A statistical watermark can be encoded in token or word choices rather than hidden characters. Deleting a zero-width character does not establish that a statistical signal changed, and Watermark Slayer does not claim a detector result.
Signed provenance and file metadata
Signed provenance, file metadata, and plain-text code points are different mechanisms. The launch scanner may report a candidate text wrapper as detected but unverified; it does not validate a signature and leaves provenance unchanged.
A reproducible check
Paste text you are authorized to inspect, run the local scan, and compare each visible finding with the supported-artifact registry. The result names the code point or structure, fixture date, support status, and available action so the claim can be checked.
Frequently asked questions
Does Claude add a watermark to text?
The phrase Claude watermark can refer to hidden characters, statistical wording signals, or signed provenance. Search interest and third-party claims do not prove that every Claude response contains the same mechanism. Check current first-party documentation and the evidence for the exact output you are inspecting.
What can Watermark Slayer remove from Claude text?
Watermark Slayer can offer reviewable cleanup only for artifacts marked supported in its public, fixture-tested registry. Other observations remain report-only or unsupported.
Does cleanup change an AI-detector result?
No detector outcome is promised. A successful cleanup confirms only the documented artifact edits shown in the review; it does not establish authorship or how another service will classify the text.