Generative AI and Lexical Shifts in Turkish Medical Publishing (2010–2025)

INTRODUCTION: The adoption of Large Language Models (LLMs) raises concerns about scientific writing integrity, affecting clinical guidelines and patient safety. While studies track AI-generated text in English databases, impacts on bilingual ecosystems remain underexplored. This study investigates the "Linguistic Equalizer" hypothesis by analyzing lexical shifts in TR Dizin, determining if LLMs are primarily utilized for content generation or linguistic refinement.
METHODS: We conducted a longitudinal bibliometric analysis of 591,819 abstracts (2010–2025), divided into "General" and "Medical" datasets. We monitored five LLM marker word groups in English and their Turkish equivalents. We evaluated statistical significance via a large-sample pooled two-proportion test (primary analysis, using risk ratios [RR] with 95% CI), a yearly-aggregate Mann-Whitney U test (secondary/directional), and segmented regression analysis to identify post-2022 trend slopes.
RESULTS: A divergence emerged. In the General Dataset, English markers exhibited a marked increase ("underscore" rose from 0.02% to 1.80% of abstracts; RR=78.73, 95% CI 64.49–96.10), while the Turkish equivalent ("vurgula") showed a more modest, though substantial, increase (RR=2.54, 95% CI 2.47–2.61). In the Medical Dataset, English "underscore" rose more sharply (RR=87.98, 95% CI 62.51–123.84), while Turkish usage showed a moderate increase (RR=2.02, 95% CI 1.90–2.15).
DISCUSSION AND CONCLUSION: The disproportionate rise in English AI markers alongside comparatively stable Turkish vocabulary is consistent with the possibility that researchers utilize LLMs primarily as translation or editing tools. These findings lend empirical support to GenAI's potential role as a 'Linguistic Equalizer,' consistent with strategic adaptation to global publishing pressures rather than content fabrication.

Keywords: Research Integrity, Artificial Intelligence, Bibliometrics, Authorship, Linguistic Equity