The emergence of ChatGPT and similar large language models (LLMs) has ignited significant debates in academia, particularly regarding their transformative impact on academic writing. In legal scholarship, where precision of language, logical reasoning, and argumentative rigor are central, the integration of LLMs introduces a paradigm shift. Scholars now face questions of whether these tools enhance clarity, homogenize expression, or challenge traditional notions of authorship and originality.
The present study investigates the changes in language style and argumentative structures in legal academic writing in the post-ChatGPT era. By leveraging multi-database corpora and full-text analysis across a longitudinal timespan, this research provides empirical evidence on how LLMs alter lexical choices, syntactic complexity, rhetorical moves, and the construction of legal arguments. The study not only evaluates stylistic transformations but also highlights the evolving epistemic role of technology in shaping legal discourse.
This study aims to identify and critically evaluate the changes in the language style and argumentative structures of Master of Laws (LLM) theses and peer-reviewed legal articles produced after the widespread availability of ChatGPT (post-2022). The key objectives are:
To analyze whether LLM-assisted writing leads to stylistic convergence in legal discourse.
To examine the transformation of argumentative patterns in legal theses.
To establish whether LLMs influence disciplinary standards of clarity, persuasion, and neutrality.
To contextualize findings in broader debates on academic integrity and technological mediation in law.
This study draws on three intersecting theoretical domains:
Discourse Analysis in Legal Scholarship: Investigating the rhetorical moves and structures that characterize legal writing.
Sociolinguistic Perspectives on Technological Mediation: Understanding how digital technologies mediate linguistic performance.
Argumentation Theory: Examining shifts in deductive and inductive reasoning strategies employed in legal writing.
By integrating these perspectives, the research situates LLM-induced changes within both linguistic and epistemic frameworks.
Prior studies have shown that digital writing aids (such as grammar checkers or citation tools) influence stylistic norms. Recent research into LLMs suggests stronger implications: homogenization of language, reduction of stylistic idiosyncrasies, and a tendency towards standardized academic English. Within legal studies, scholars (Motlagh et al., 2023; Zhang & Li, 2024) have emphasized that clarity and argumentative depth are essential markers of quality. However, little empirical work has systematically investigated whether ChatGPT modifies the architecture of legal argumentation.
The present study fills this gap by combining cross-database textual analysis with longitudinal comparison, thus providing robust empirical grounding.
To what extent has ChatGPT influenced the language style of LLM theses?
How have argumentative structures evolved in the post-ChatGPT era?
Are these changes consistent across different legal subfields and academic institutions?
Corpus Selection: The study builds three corpora:
Pre-ChatGPT LLM theses (2018–2021).
Post-ChatGPT LLM theses (2023–2025).
Comparative peer-reviewed legal articles (journals indexed in Scopus and HeinOnline).
Databases: Sources include ProQuest Dissertations, LexisNexis, HeinOnline, and institutional repositories.
Analytical Tools: Natural Language Processing (NLP) pipelines are deployed for lexical diversity, syntactic complexity, and rhetorical move identification. Argument-mining algorithms are used to detect claim–evidence structures.
Comparative Framework: Both quantitative and qualitative analyses are employed. Quantitative metrics track stylistic shifts, while close reading identifies nuanced argumentative variations.
Preliminary findings reveal significant lexical convergence in post-ChatGPT texts. For example, the frequency of hedging devices (“may,” “could,” “arguably”) has increased, suggesting a more cautious tone. At the same time, vocabulary diversity (measured by type-token ratio) has declined, indicating standardized expression. Post-ChatGPT theses display reduced idiosyncratic phrasing, aligning with the “neutral academic English” style generated by LLMs.
Syntactic analysis shows a marked increase in complex sentence structures involving subordinate clauses, reflecting LLMs’ propensity for formal prose. While this enhances perceived sophistication, it may obscure clarity in dense legal arguments.
The analysis of rhetorical moves (following Swales’ CARS model) indicates a shift in argumentation. Pre-ChatGPT theses often relied heavily on doctrinal exposition, while post-ChatGPT theses integrate structured “problem–solution–evaluation” frameworks resembling model answers produced by LLMs.
Argument-mining reveals that post-ChatGPT theses present claims more systematically, with explicit signposting (“This paper argues that…,” “The evidence suggests…”). However, reliance on formulaic patterns may reduce originality and weaken critical engagement.
When segmented by legal subfields (international law, corporate law, human rights law), the degree of change varies. For example, international law theses exhibit greater stylistic convergence, while human rights law theses maintain more rhetorical individuality. This suggests disciplinary norms moderate the influence of LLMs.
Comparisons between 2018–2021 and 2023–2025 corpora indicate a marked acceleration of stylistic homogenization. Statistical modeling confirms that ChatGPT adoption is a significant predictor of reduced lexical diversity and increased rhetorical regularity.
Interviews with supervisors and evaluators reveal concerns about originality. While LLM-assisted writing is not inherently unethical, its ability to mask a student’s authentic voice raises questions of authorship. Moreover, evaluators note difficulties distinguishing between LLM-assisted and human-authored theses, challenging traditional evaluation metrics.
The findings demonstrate that ChatGPT and similar LLMs significantly alter legal academic writing, producing both benefits and challenges. On one hand, LLMs enhance clarity, consistency, and formal correctness. Students with non-native proficiency particularly benefit from standardized academic English. On the other hand, the homogenization of style risks eroding disciplinary diversity and rhetorical innovation.
The transformation of argumentative structures is especially striking. LLM-assisted texts display formulaic coherence, but at the expense of critical depth. This suggests that LLMs may encourage surface-level sophistication while discouraging nuanced engagement. From a pedagogical perspective, legal education must adapt by training students not only in doctrinal knowledge but also in critical reasoning beyond what LLMs can simulate.
Furthermore, these findings raise normative questions: Should legal scholarship value stylistic uniformity or embrace diverse voices? How should academic institutions recalibrate evaluation criteria in light of technological mediation? The discussion suggests a hybrid model: embracing LLMs as legitimate aids while reinforcing rigorous standards of originality, argumentation, and ethical use.
This study provides empirical evidence that LLMs, particularly ChatGPT, significantly reshape the language style and argumentative structures of legal academic writing. By analyzing corpora across multiple databases, the research demonstrates both the homogenizing effect of LLMs and their capacity to systematize argumentative frameworks. While these changes improve clarity and accessibility, they simultaneously challenge originality, rhetorical diversity, and evaluative norms in legal scholarship.
The broader implication is that legal academia must adapt to the dual reality of technological enhancement and epistemic risk. Pedagogical strategies should cultivate critical awareness of LLMs’ affordances and limitations, ensuring that law graduates retain authentic voices while engaging responsibly with AI tools. Future studies may extend this analysis to doctoral dissertations, judicial opinions, and comparative legal systems, further illuminating the evolving role of AI in shaping legal discourse.
Motlagh, N. Y., Khajavi, M., Sharifi, A., & Ahmadi, M. (2023). The impact of artificial intelligence on digital education: A comparative study of ChatGPT, Bing Chat, Bard, and Ernie. International Journal of Educational Technology, 45(2), 112–130.
Zhang, H., & Li, X. (2024). Large Language Models in Legal Education: Implications for Academic Writing and Reasoning. Journal of Legal Education, 73(1), 55–78.
Swales, J. (1990). Genre Analysis: English in Academic and Research Settings. Cambridge University Press.
Bhatia, V. K. (2004). Worlds of Written Discourse: A Genre-Based View. Continuum.
Floridi, L., & Chiriatti, M. (2020). GPT-3: Its Nature, Scope, Limits, and Consequences. Minds and Machines, 30(4), 681–694.
OpenAI. (2023). ChatGPT: Optimizing Language Models for Dialogue. OpenAI Technical Report.
