The influence of generative artificial intelligence on the stylistic features of a digital text

Abstract

The spread of large language models has changed the conditions under which written text is produced in the digital environment and has confronted stylistics with the question of which features distinguish machine generation from the writing of a native speaker. The digital text, previously described mainly through the technical parameters of its medium, has acquired an additional source of variation – the algorithmic mechanism of production, which leaves a stable stylistic trace on the surface of speech. Using a model corpus of 612 texts of four genres covering the period 2023–2026, the study compares three types of digital text: the author-written text, the fully generated text and the generated text with subsequent editing. The corpus method, componential analysis of meaning and stylostatistical processing of quantitative parameters are applied. It is established that a generative model narrows the dispersion of sentence length by roughly 1.6 times, raises the density of discourse connectives to 16.8 per thousand word tokens and brings the share of summarising paragraph endings up to 63.4 %, while lexical diversity falls by 15.7 %. Manual editing weakens most of the markers, yet along certain syntactic parameters it reinforces them. The resulting distribution shows that the stylistic trace of generation is formed predominantly at the compositional and syntactic levels and is poorly removed by cosmetic correction; it follows that the diagnosis of the machine origin of a text relies primarily on structural rather than lexical characteristics.