Prompt-Conditioned Variation in English Academic Texts: A Stylistic and Pragmatic Analysis of AI-Generated Writing

Abstract

This research examines the impact of prompting on linguistic/style/pragmatic features in English academic writing samples produced by AI. Instead of adopting a presupposition of stability for the category text produced by AI, the paper theorizes this prompt-based discourse as: an emergent textual product shaped by prompt wordings, tone, disciplinary positioning, rhetorical purpose and level of authorial commitment. The research constructs a controlled prompt matrix with twelve prompt conditions and an advised final corpus of 120 academic texts produced in clear, repeatable circumstances. The analysis is intentionally empirical lexical complexity, hedging, density of modality, density of stance, depth of argument index, density of transitions, mean sentence length and coherence index are computed through corpus-linguistic and pragmatic analyses. The pilot statistical model indicates differentiation results of critical, cautious, methodological, high-quality and pragmatic-reflexive prompts in terms of academia profile. Pragmatic-reflexive and critical prompts strengthen stance and argument depth; cautious prompts strengthen hedging and modality; high-quality and methodological academic prompts strengthen lexical complexity; concise prompts weaken argument depth as well as pragmatic complexity. The key contribution is the introduction of the Prompt-Conditioned Variation Framework (PCVF) that conceptualizes AI writing as an interactional, instruction sensitive register, rather than a homogenous machine signature. The implications are for research in English language writing, assessment with the writing task, academic integrity and prompt literacy in 2026 through 2036.