Abstract
This study examines the linguistic features of legal responses generated by artificial intelligence on the LegalExpert.kz platform. The aim of the research is to identify structural patterns at the lexical, syntactic, and pragmatic levels of such responses. The dataset consists of 50 user queries on actual legal issues in Kazakhstan and the corresponding AI-generated answers. The findings demonstrate that, at the lexical level, legal terms such as “court,” “case,” “evidence” and “document” occur with high frequency and form the semantic core of the texts. At the syntactic level, responses are structured according to predefined templates, with a noticeable predominance of conditional constructions (e.g., “if… then…”). This reflects the linguistic realization of the legal reasoning model “situation–condition–consequence”. At the pragmatic level, the responses are user-oriented, action-guiding, and express modality related to probability and caution. In addition, the results reveal a blending of legal discourse with everyday language. This indicates that AI-generated texts are adapted to make complex legal information accessible to a broad audience. The study contributes to understanding the linguistic organization of legal AI systems and provides insights into evaluating their communicative effectiveness.