JMIR Publications

ABSTRACT

Abstract Generative artificial intelligence is increasingly used in medical manuscript preparation. Large language models may improve readability, organization, and access to scientific communication, particularly for authors who are not native English speakers. The same technologies may also create new vulnerabilities for research integrity by generating fabricated or misused references, persuasive but unsupported interpretation, and polished manuscripts that conceal weak evidence or inadequate human intellectual contribution. This position paper analyzes generative AI in medical publishing as a challenge for artificial intelligence in medicine, because unreliable manuscripts may enter evidence synthesis, clinical guidelines, educational materials, and ultimately clinical decision making. We propose a three-group framework: honest researchers who use AI to improve communication; dishonest researchers who use AI to accelerate deception; and an intermediate group of fundamentally honest authors who may adopt shortcuts under academic or financial pressure. The central risk is not visibly artificial text but credible-looking manuscripts whose form imitates scholarship while their evidentiary substance remains weak or unverified. We recommend transparent AI disclosure, manual reference verification, data availability, reviewer education, editorial screening for integrity risks, and reform of promotion systems that reward publication volume over scientific quality.