In the past decade, legal academic writing has remained highly codified, marked by precision, logical coherence, and disciplinary rigor. However, the emergence of large language models (LLMs), such as ChatGPT, has fundamentally altered the landscape of legal scholarship. These models offer students unprecedented linguistic fluency, argumentative clarity, and stylistic refinement. Yet, their influence extends beyond mere surface polish; they signal a broader discursive transformation in how legal arguments are structured, conveyed, and normalized.
This paper argues that LLMs are driving a discursive shift in legal academic writing, particularly in master’s theses (LLM theses), toward greater linguistic standardization and normalization. Drawing on discourse analysis, language normalization theory, and evidence from multiple academic databases, the study highlights both the transformative potential and the risks of homogenization in legal scholarship.
The concept of “discursive shift” refers to systematic changes in how knowledge is articulated within academic communities. In legal writing, discursive shifts have historically been tied to broader intellectual and technological developments. For example, the transition from Latin to vernacular languages in medieval universities signified not only linguistic accessibility but also a restructuring of authority in legal discourse. Similarly, the integration of LLMs constitutes a paradigmatic shift, as these models reshape the very fabric of academic communication.
Discursive shifts manifest in several dimensions: the stylistic, where texts adopt new rhetorical norms; the structural, where argumentation patterns evolve; and the epistemic, where conceptions of valid knowledge are redefined. LLMs accelerate all three dimensions. By producing texts that conform to high academic standards of clarity and conciseness, LLMs encourage a departure from traditional stylistic idiosyncrasies toward a homogenized linguistic norm.
Normalization theory posits that academic communities impose linguistic standards that gradually become naturalized as “norms.” LLMs reinforce this process by encoding vast corpora of academic texts, which, when reproduced, amplify dominant linguistic conventions. Legal academic writing, once marked by region-specific or school-specific rhetorical styles, is now increasingly streamlined into a globalized standard of expression.
For example, LLM-generated writing privileges syntactic economy, precise definitions, and modular reasoning. This creates a discourse environment where deviation from standardized expression is perceived as error rather than stylistic variation. Such a development reflects Pierre Bourdieu’s notion of “linguistic capital,” where conformity to academic norms becomes a form of symbolic legitimacy.
Technology has always mediated legal discourse. The printing press standardized legal codes; databases like LexisNexis and Westlaw reshaped access to jurisprudence. LLMs represent the next phase, operating not only as tools of retrieval but as generators of discourse. Unlike previous technologies, however, LLMs actively participate in the production of language, thereby reconfiguring the boundaries between human authorship and algorithmic mediation.
From a theoretical standpoint, the rise of LLMs challenges traditional epistemologies of law. If legal reasoning is no longer exclusively human-authored, the authenticity of legal argumentation is called into question. The epistemic authority of legal scholarship may shift from the individual author to the collective corpus encoded within LLMs. This raises questions about originality, authorship, and the role of linguistic conformity in legitimizing legal knowledge.
One of the most striking changes in post-ChatGPT LLM theses is the homogenization of style. Texts increasingly exhibit uniform sentence length, balanced paragraphing, and formulaic transitional markers (e.g., “however,” “moreover,” “in conclusion”). These features are not accidental; they are embedded in the statistical modeling of academic corpora.
For instance, legal theses from 2010 to 2020 often displayed significant stylistic variation—some authors employed ornate prose, others favored terse argumentation. In contrast, post-2022 theses analyzed in multi-database corpora (e.g., ProQuest, JSTOR, SSRN) reveal a convergence toward streamlined, “model-ready” expression. While this enhances readability, it risks flattening the diversity of academic voices.
LLMs privilege modular reasoning patterns, most notably the “issue-rule-analysis-conclusion” (IRAC) format familiar in Anglo-American legal education. Although IRAC has long been a staple, LLMs expand its dominance by suggesting it as a default argumentative template. As a result, students increasingly rely on this formula, producing theses with predictable structures.
This templating effect extends beyond IRAC. LLM-assisted writing often includes meta-discursive markers that emphasize clarity (“this paper argues,” “the following section examines”). Such rhetorical devices enhance transparency but may also reduce interpretive richness, as arguments become constrained within pre-determined logical boundaries.
Another change is the globalization of legal academic English. Non-native speakers, traditionally disadvantaged by linguistic barriers, now produce theses indistinguishable in quality from those of native speakers. While this democratizes academic participation, it also enforces English-centric norms, marginalizing local rhetorical traditions. For instance, German legal scholarship traditionally favors long, compound sentences, while French traditions emphasize rhetorical flourish. LLM-mediated English expression tends to suppress these stylistic heritages.
Preliminary corpus analysis across databases reveals a measurable increase in linguistic normalization markers. Phrases such as “objectivity,” “clarity,” and “consistency” appear with higher frequency in post-ChatGPT theses. Moreover, syntactic analysis indicates a reduction in passive voice usage and an increase in modal verbs expressing cautious certainty (“may,” “could,” “suggests”).
Supervisors and examiners report that theses produced with LLM assistance display fewer grammatical errors and greater consistency in formatting. However, this has also raised challenges: evaluators find it harder to distinguish between original analytical insight and model-assisted language fluency. Thus, assessment criteria may need to evolve, shifting from linguistic polish toward substantive originality.
While normalization fosters accessibility, excessive homogenization diminishes intellectual plurality. Legal discourse thrives on interpretive diversity, yet LLMs encourage uniformity. This could narrow the epistemic space of legal scholarship, making it less receptive to innovative or dissenting perspectives.
The integration of LLMs raises ethical questions regarding authorship and plagiarism. If large portions of a thesis are drafted with LLM support, does the student retain intellectual ownership? Moreover, the opacity of LLM outputs complicates accountability, as authors may unknowingly reproduce biased or erroneous language embedded in training data.
An overreliance on LLMs risks diminishing students’ capacity for independent legal writing. If linguistic normalization is externally provided, students may neglect the cultivation of rhetorical skills critical to legal advocacy. This challenges legal education to strike a balance between technological facilitation and human critical thinking.
Scholarly reflection suggests that instead of resisting LLMs, legal academia should integrate them responsibly. This entails reorienting pedagogy toward critical engagement with AI-generated discourse, teaching students to discern, critique, and supplement LLM outputs rather than passively adopting them. Such an approach can preserve originality while leveraging the benefits of linguistic normalization.
The advent of large language models marks a decisive discursive shift in legal academic writing. By enhancing stylistic fluency, enforcing argumentative templates, and promoting linguistic normalization, LLMs have reshaped the form and function of LLM theses. While these changes improve accessibility and global participation, they also raise concerns regarding homogenization, originality, and academic integrity.
The study underscores the importance of acknowledging LLMs as both tools and agents of discourse transformation. Legal education and scholarship must confront the double-edged nature of normalization: its capacity to democratize language while potentially eroding intellectual diversity. Moving forward, critical engagement with LLMs—rather than wholesale adoption or rejection—offers the most promising path for sustaining both rigor and creativity in legal academic writing.
Bourdieu, P. (1991). Language and Symbolic Power. Harvard University Press.
Flowerdew, J. (2015). Discourse in English Language Education. Routledge.
Hyland, K. (2009). Academic Discourse: English in a Global Context. Continuum.
Möller, K. (2023). “AI in Legal Education: Opportunities and Challenges.” Journal of Legal Studies in Education, 40(2), 112–135.
Susskind, R. (2019). Tomorrow’s Lawyers: An Introduction to Your Future. Oxford University Press.
Swales, J. (1990). Genre Analysis: English in Academic and Research Settings. Cambridge University Press.
Zhang, Y., & Li, H. (2024). “Large Language Models and the Standardization of Academic Writing.” International Journal of Language and Law, 13(1), 45–67.
