Who gets seen in the age of AI? Adoption patterns of large language models in scholarly writing and citation outcomes

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

The rapid diffusion of generative AI tools is reshaping how scholars produce and communicate knowledge, raising questions about who benefits and who is left behind. We analyze over 230,000 Scopus-indexed computer science articles (2021–2025) to describe how stylistic markers of AI-likeness in scholarly writing co-evolve with scholarly visibility across regions. Using a zero-shot detector (Binoculars) as a proxy for AI-likeness, we document stylistic changes in writing and examine their associations with citation counts, journal placement, and global citation flows in the periods before and after the public release of ChatGPT on November 30, 2022. We emphasize that the Binoculars score captures stylistic similarity to large-language-model outputs rather than confirmed AI usage, and that all reported relationships are associational rather than causal. We document several patterns: stylistic AI-likeness rises across regions after late 2022 (in the detector score, in its binary classification, and in a detector-independent lexical marker index), with a more pronounced shift among authors in the Global East; AI-like writing is associated with higher citation counts conditional on venue and journal-prominence characteristics, an association that persists when generative-AI-topic papers are controlled for or excluded and replicates under a second neural detector, though it attenuates under the coarser lexical measure; more human-like writing is conversely associated with placement in higher-tier journals; and global citation networks show a modest eastward tilt in visibility alongside tighter intra-regional clustering, with country-level changes in AI-likeness unrelated to changes in network position. These patterns are consistent with a co-occurring reconfiguration of scholarly writing and recognition around the diffusion of generative AI, but we cannot rule out concurrent topic shifts, productivity effects, and the broader rise of generative-AI research within computer science as alternative explanations.

Links and sources

Comments