The disclosure problem
The question a standards desk actually asks is narrow: was any part of this produced by a system that marks its output, and can we show what we checked? That is answerable. “Does this feel like AI” is not.
WatermarkRemoverPro answers the narrow question and refuses the broad one. The result names the keys tested and states plainly that a null result under those keys is not a statement about marks applied with a key nobody outside the vendor holds.
Why the per-passage breakdown is corrected
A 1,500-word piece contains dozens of passages. Testing each one separately and reporting whatever looks significant would light up two or three sentences of anybody’s writing, every time. That is what running many simultaneous tests does.
We apply a false-discovery-rate correction across all passages before presenting any of them as a finding, and the report says how many were tested and how many survived. An uncorrected highlighter is a machine for generating false accusations inside your own newsroom.
Source material and confidentiality
The free check runs entirely in the browser, so unpublished copy is not transmitted to us. For a newsroom that is often the deciding factor: the alternative is uploading an embargoed piece to a third-party service.
If you use the API or the report endpoint, those necessarily run on our servers, and that is documented rather than buried, so you can make that decision knowingly.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. WatermarkRemoverPro's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.