WatermarkRemoverPro

For university students accused of using AI

Being told your essay “came back as AI” is frightening partly because the number you are shown usually arrives with no explanation and no way to interrogate it. Before you write anything to your department, it is worth knowing what a statistical test can and cannot establish about a document, including this one.

What the number your university showed you probably is

Most institutional tools report a percentage that sounds like “this much of the document is AI”. It almost never means that. These systems are classifiers trained to separate two corpora, and the figure is a model’s confidence, not a proportion of your text and not a probability that you personally used a model.

That distinction matters enormously in an appeal, because a confidence score has a false positive rate, and a false positive rate applied across a whole cohort produces a predictable number of wrongly accused students every term. You may be one of them, and the arithmetic is on your side more than it feels.

What WatermarkRemoverPro checks instead

WatermarkRemoverPro looks for a provenance mark: a deliberate statistical signature that some generation systems apply to their output so it can later be recognised. That is a different question from “does this read like AI”, and it has a cleaner answer, because a mark is either present in the statistics or it is not.

The honest catch, which we state on every result: a mark of this kind is keyed, and no model vendor publishes its detection key. So a null result from us means “no mark found under the keys we hold”, never “you are cleared”. We would rather give you a narrow true statement you can defend than a broad one that collapses the moment someone knowledgeable reads it.

Putting an appeal together

Ask the department three things in writing: which tool produced the figure, what that tool’s published false positive rate is, and what corroborating evidence exists beyond the score. Many academic misconduct policies already require more than a detector output, and asking politely for the policy tends to be more productive than arguing about the number.

Bring your own process evidence: drafts, version history, notes, library loans, supervision emails. A document history is far more persuasive than any detector result in either direction.

The WatermarkRemoverPro evidence report is designed to sit alongside that: it is dated, states its own limits in full on the page, lists the keys tested, and carries a SHA-256 hash of the exact file so it cannot be argued to be about a different draft.

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.

Answers, in full

Questions people actually ask

Will a WatermarkRemoverPro report clear my name?
It cannot, and any tool promising that is misleading you. No absence of evidence proves authorship. What the report does is document what a specific, named statistical test found on your exact file, with its limits stated, so the conversation moves from an unexplained percentage to something both sides can examine.
Is my essay uploaded anywhere?
Not on the free check. The analysis runs in your browser, and you can open the network tab and watch. That matters here specifically: uploading an unsubmitted essay to a third party is itself something some departments take a dim view of.
I did use AI for some parts. What now?
WatermarkRemoverPro's rewrite feature is a final-pass editing tool for your own drafting, not a substitute for disclosure your institution requires. It cannot guarantee a document won't be flagged, and it doesn't change what happened to a document that already went through an assisted-writing process your institution's policy required you to disclose. If your institution permits assisted writing with disclosure, disclose it. If it does not, an honest conversation earlier is almost always treated better than a discovered concealment later.