Skip to content
WatermarkAudit

Manuscript screening

Screen a submission. Keep the record.

Drop in a submission package and see what it declares, what its files carry, and whether those two things agree. Every finding separates what was observed from what it means and what to do — because an observation written in the voice of a verdict is how screening tools get authors into trouble they don't deserve.

This is not an AI detector, and no honest tool is one. It reads declarations, credentials and metadata, and tells you what is there.

Drop a submission package

or

A zip, or the files themselves. Nothing is uploaded — every check runs in this browser.

What the screen checks

Disclosure

Finds the AI-use statement under its own heading — the Elsevier and Frontiers wordings and their relatives — and reads whether it declares use, declares non-use, or says neither. Text under no such heading is discussion, not declaration, and is left alone.

Provenance

Verifies any content credential on every file in the package, and reads the metadata that survives when a credential does not.

Consistency

Compares the declaration against the technical evidence, in both directions. It confirms agreement as clearly as it flags a contradiction.

Coverage

Where files carry AI provenance, whether the manuscript's disclosure accounts for them. A statement about language editing does not account for a generated figure.

The package

Duplicate files, mixed capture devices, and provenance that is present on some files and absent on others — questions only visible across a set.

Other statements

Ethics, funding, data availability, competing interests and author contributions, reported as present or not found. Whether one is adequate is an editorial judgement, not ours.

What it deliberately does not do

  • It does not estimate whether text was written by AI. Those classifiers are unreliable and fall hardest on authors writing in a second language.
  • It does not judge images for manipulation. The available techniques produce false accusations at a rate no editor should act on.
  • It does not claim a submission is compliant with anything. It documents what was checked, which is the part a journal can actually stand behind.

For a pilot journal

Screening works without an account. Signing in adds the part a journal needs: a manuscript reference, a dated history of every screen and editorial decision, and an append-only record you can produce if a submission is ever challenged.

Running a journal and want to try this properly? Get in touch — a pilot is exactly what this needs.