Where publishers already use it
Submission screening. Automated checks for scope, formatting, reference integrity, duplicate submission and plagiarism run before a human editor sees a manuscript.
Integrity checking. Image manipulation and duplication detection, statistical implausibility checks, and detection of paper-mill signatures such as tortured phrasing and manipulated citation patterns.
Reviewer matching. Suggesting reviewers by topic and conflict screening.
Production. Structuring, tagging, reference linking, and format conversion — the least contentious application.
Discovery. Recommendation and search on publisher and platform sites.
Where the gain is clearest
Research integrity. Paper mills produce fabricated manuscripts at a volume human editorial staff cannot screen, and image-duplication detection finds problems that defeated manual review for years. This is work that was not being done rather than work being replaced.
Where it is contested
AI in peer review. Several funders and publishers prohibit uploading manuscripts under review to general AI tools, on confidentiality grounds, and some prohibit AI-generated reviews outright. The objection is both confidentiality and that a review’s value lies in a qualified human’s judgement.
Detection of AI-written text. Detectors are unreliable, with false positives that fall disproportionately on non-native English writers. Using a detector score as evidence of misconduct is not defensible on current performance.
Editorial decisions. Desk rejection influenced by automated scoring raises questions of appeal and transparency that are not settled.
What authors should know
Your manuscript is probably being screened automatically before an editor reads it, which is a reason for clean formatting, resolvable references and a scope-appropriate target.
If accused on the basis of a detector, ask what the evidence is. Detector output alone is weak evidence and should be challenged as such.
In short
What to take from this.
- Expect: automated screening before human editorial review.
- Treat AI-text detector scores as weak evidence, not proof.
- Note: many publishers and funders prohibit AI tools in peer review.
Sources
- COPE guidance on artificial intelligence in decision-making
- STM integrity initiatives on paper mills and image screening
- Published evaluations of AI-text detector accuracy
Related