Linguistic shifts in dental scientific writing in the large language model era.

Abstract

Background: The integrity of scientific literature depends on transparent authorship and authentic communication of research findings. The release of ChatGPT in November 2022 introduced powerful AI writing assistance that may be changing how dental researchers write manuscripts. However, systematic evidence documenting this shift and its implications for research integrity remains limited. Traditional detection tools and self-reported surveys have proven unreliable, requiring alternative approaches to assess AI's influence on published literature.

Objective: To detect and quantify linguistic changes in dental scientific abstracts by comparing pre-ChatGPT (2021) and post-ChatGPT (2025) periods, providing empirical evidence of AI adoption patterns and their implications for scientific communication integrity.

Methods: We analyzed 2,770 abstracts from six major dental journals using vocabulary-based linguistic analysis. Following computational linguistics methods validated in biomedical literature, we extracted six AI-associated features and computed composite detection scores. This observational text-based approach avoids reporting bias while providing quantifiable evidence of linguistic shifts.

Results: AI detection scores increased significantly across all journals (mean: 12.01 to 14.46; + 20.4%), with individual journal increases ranging from +10.0% to +34.5%. The mean number of AI marker words per abstract rose from 0.69 to 0.97 (+41.1%), while high-suspicion abstracts (score ≥ 20) increased from 18.0% to 25.8% (+43%). Changes were consistent across journal specialty, geography, and impact factor, suggesting field-wide adoption.

Conclusions: Dental scientific writing has undergone measurable linguistic changes coinciding with widespread AI availability. The consistency and magnitude of these changes raise questions about transparency in AI use, maintaining authorial voice, and preserving the integrity of scientific communication. These findings provide an empirical foundation for evidence-based policies on AI disclosure and appropriate use in academic publishing, grounded in observed changes to scientific communication rather than speculative concern.