Research integrity screening is software that checks each paper submitted to a journal for signs of fraud, before the paper is sent for peer review. It looks for copied text, reused or edited images, invented data, papers bought from paper mills, and writing done by AI that the authors did not declare. A paper mill is a business that sells fake research papers, or places on the author list of a paper, to people who need publications for a degree, a job or a promotion.
Why screen before peer review? Peer reviewers are experts who read papers without pay, in their spare time. They check whether the science makes sense. They are not trained to spot a reused photograph, or a paper written by a paper mill. If such a paper is published and found out later, the journal must retract it, which means formally withdrawing it. Screening first saves the reviewers' time and protects the journal's record.
The software runs several checks. A text similarity check compares the paper with millions of published papers and web pages, and reports what percentage of the text matches. A tortured-phrase check catches a common trick. Some fraudsters copy a paper and pass it through paraphrasing software, which replaces technical terms with odd synonyms. For instance, "artificial intelligence" can become "counterfeit consciousness". An image check looks for figures that were taken from another paper, then flipped, cropped or edited. A reference check looks for references that do not exist, and for cited papers that have already been retracted.
Here is a typical case. The editorial office of a medical journal in Guwahati receives a paper on a diabetes drug. The screening report shows 14% similar text, which is normal for a methods section. But it also finds three tortured phrases. One microscope image matches an image in a 2021 paper by different authors, rotated by 90 degrees. Seven of the 45 references cannot be found in any database. The software does not reject the paper. It flags it, with the evidence, and an editor decides. Here, the editor rejects the paper and writes to the authors' institution.
Detecting writing done by AI is the least reliable check. Detectors often mark honest writing by non-native speakers as AI writing. So editors treat that score only as a reason to look closer. Checking whether an image was generated or edited uses methods similar to deepfake detection. Fakes made with AI are discussed in AI-generated content.
The reference check resembles a citator in law. A citator tells a lawyer whether a judgment has since been overruled. The reference check tells an editor whether a cited paper has since been retracted.
Indian universities run similar checks on theses. Under the University Grants Commission's 2018 rules, a thesis with more than 10% similar text faces penalties that grow with the percentage. Journals run their screening as one step inside the journal publishing platform.