Screen what comes in, and what went out
Assistant fingerprints, excess vocabulary, tortured phrases, DOIs that belong to another paper, cited works that were retracted, reused or AI-declared pictures — with the evidence, for submissions, for your recent output as the public record shows it, and for reviewer reports that look alike.
The shared duplicate-submission check
Journals that take part find out when a submission they screen is very like one another journal taking part screened in the last 12 months — the same manuscript sent to two journals at once. Each journal decides for itself, here; it costs nothing.
Sign in (free) to choose for your journals.
What is kept, for how long, who sees what
What is kept. For each manuscript your journal screens while it takes part: one fingerprint — 128 numbers worked out from the manuscript's five-word phrases (a MinHash signature), filed under 32 short keys made from those numbers so that a match is found quickly — your journal's number on this site, and the date. No text, no title, no names, no file. The numbers cannot be turned back into words. The reference list is left out.
For how long. 12 months. A fingerprint is deleted once it is 12 months old — the site's privacy clock looks twice an hour — and when your journal leaves, every fingerprint it added is deleted at once.
Who sees what. Only journals taking part, and only this, in the report of their own screening: "A manuscript very like this one was screened by another journal taking part, on date" — about 60% or more of the five-word phrases in common. Never which journal, never the other manuscript, never its authors. No author, index or other site sees anything.
What a match means. That the same text went to two journals within a year. Authors may have withdrawn it first, or been rejected; a match is a question to put to them, never a finding.
Use it as a policy, not a verdict
A strong signal is a reason for a person to look, never a rejection on its own: ask the author to explain the flagged passages, to supply original figures, to verify the flagged references. Write that into your ethics page: "submissions are screened for AI-writing, paper-mill and image signals; findings are put to the authors before any decision". The Journal Audit now checks for exactly that statement, for an AI-use policy and for an image-integrity policy.
Strong needs two independent families of signals; one family alone reads Notable at most, and the vocabulary is not counted when the authors declare language editing, translation, or English as an additional language (from 9 Oct 2026).
Screening at submission, by your own script
There is no ready-made OJS plugin for this yet. With the institution or partner API key, a script of your own (or your developer's, hooked to OJS's submission step) can post each new submission's text and get the same signals back as JSON: https://partners.smartscholars.in/api.php?action=screen. The API page shows the request and the answer.
What it cannot tell you
Whether a clean-looking text was written by a person — no tool can, and the ones that claim to misfire on technical English. Whether a figure that carries no metadata was generated. Whether a reviewer is real. It tells you what the text, the references and the files themselves show, with the rule and the source, so that your decision is yours and can be explained to the author.
