Human-Written and AI-Generated Research Article Abstracts:A Paired Rhetorical Move Analysis of Indonesian ELT and Linguistics Articles
TL;DR
Gemini can produce structurally conventional research abstracts, but its rhetorical organization tends to be more regularized and less variable than that of human writers, which suggests that Indonesian researchers and students may benefit from using Gemini as a supporting tool for checking rhetorical completeness and clarity.
Abstract
This study compares the rhetorical organization and textual characteristics of research article abstracts written by Indonesian-affiliated authors with abstracts generated by Gemini 3.1 Pro from the same source articles. Using a paired small-corpus design, 20 human-written abstracts from four Indonesian ELT and linguistics journals were compared with 20 AI-generated counterparts. Rhetorical moves were identified using Swales and Feak’s five-move framework, and word and sentence counts were also examined. The results show that Purpose, Methodology, and Results occurred in all abstracts in both corpora, forming a stable rhetorical core. Human-written abstracts included Background more frequently, whereas Gemini consistently included a Conclusion and used fewer move-sequence patterns. Human writers also devoted more sentence space to Background and Methodology, while Gemini consistently foregrounded Results. Overall abstract length was relatively similar, with human and AI abstracts averaging 201.95 and 208.80 words, respectively. These findings suggest that Gemini can produce structurally conventional research abstracts, but its rhetorical organization tends to be more regularized and less variable than that of human writers. Therefore, Indonesian researchers and students may benefit from using Gemini as a supporting tool for checking rhetorical completeness and clarity while retaining human control over the sequencing, emphasis, and distribution of rhetorical moves.