Variability in Journal and Publisher Policies on Artificial Intelligence Use in Manuscript Preparation: An Orthopaedic Perspective.

BACKGROUND: Artificial intelligence (AI)-generated text has been detected in nearly 90% of orthopaedic peer-reviewed publications, yet public disclosure of AI use remains extremely low. Whether journal and publisher policies for authors on AI use are present, sufficiently detailed, and internally consistent has not been systematically examined in orthopaedics. We characterized the prevalence, scope, and within-publisher consistency of AI policies across a structured sample of leading orthopaedic and biomedical journals. METHODS: We performed a cross-sectional analysis of publicly available author instructions, editorial policies, and publisher guidance for 32 journals (6 core orthopaedic, 19 orthopaedic subspecialty, 7 general biomedical) across 12 biomedical publishers. Forty-one variables were extracted per journal between March and April 2026, including presence of policy for AI use for authors, disclosure requirements, and use of AI for data analysis, generating figures and references, and policies for peer reviewers. A second reviewer independently coded a 12-variable subset for 10 journals (31% calibration) to assess inter-rater reliability (Cohen's kappa = 0.90, 96% agreement). RESULTS: Twenty-nine of 32 journals (91%) had an identifiable AI policy for authors; 3 subspecialty journals had none. All core orthopaedic (6/6) and general biomedical (7/7) journals had policies, versus 16 of 19 subspecialty journals (84%). Among journals with policies, 28 of 29 (97%) required disclosure of AI use by authors and 22 of 29 (76%) prohibited listing AI as an author. Coverage of other domains was less frequent: figures and images (66%), hallucinations or fabrication (62%), AI-generated references (48%), peer review (45%), and data analysis (41%). CONCLUSIONS: Domains that can directly impact scientific integrity and clinical practice, such as AI-assisted manipulation of data, statistical outputs, fabricated references, and undisclosed image alterations, are currently not consistently queried in detail at manuscript submission. With advances in software technology, biomedical journals can test and adopt AI-detection tools that are both sensitive and specific to screen all submissions before peer-review to maintain scientific integrity and transparency. LEVEL OF EVIDENCE: Level IV, Cross-Sectional Study.