AI-ASSISTED LEARNING MATERIALS DEVELOPMENT FOR ESP: AFFORDANCES AND LIMITATIONS

Abstract

Generative artificial intelligence has moved rapidly from a novelty to a routine part of how many English for Specific Purposes (ESP) teachers prepare their materials, yet the normative and empirical picture of how these tools should be used has not kept pace. This article examines AI-assisted learning materials development for ESP from three complementary angles: the legal, ethical, and competency-based frameworks that govern AI use in the European Union and Ukraine; the existing research literature on AI-assisted materials development; and empirical evidence from ESP practitioners themselves. The study followed a three-stage mixed-methods design: an analysis of normative documents, including the EU AI Act, Ukrainian copyright and academic integrity legislation, EU and UNESCO ethics guidelines, and digital and AI competency frameworks; a literature review of Web of Science records screened for relevance; and a questionnaire administered to 46 ESP teachers at Igor Sikorsky Kyiv Polytechnic Institute, analysed through the Technology Acceptance Model (TAM). The normative analysis shows that, despite differing legal mechanisms, EU and Ukrainian frameworks converge on requiring human oversight, transparency, and disclosure of AI-generated content. The literature review confirms that AI tools increase efficiency and personalisation in materials development while introducing risks to quality, bias, and disciplinary accuracy, a tension sharper in ESP than in general education because of the field's dependence on precise, discipline-specific content and language. The survey results show that AI tool use is near-universal among the sampled teachers (95.7%), concentrated in vocabulary, grammar, and field-specific text generation. TAM analysis confirms all four hypothesised relationships between perceived ease of use, perceived usefulness, attitude, and behavioural intention, with behavioural intention rated highest and time savings receiving the highest rating among the perceived usefulness items; notably, teachers were less positive about the time required to refine AI-generated output before classroom use. Perceived quality varied across exercise types, favouring open, communicative tasks over tightly structured ones, and self-assessed AI competency remains concentrated at the basic level, particularly for AI ethics. These findings indicate that current ESP practice is best described as AI-assisted rather than AI-driven, with teachers retaining a central evaluative role consistent with normative requirements, though a literacy gap, especially in ethics, remains the main obstacle to confident AI-assisted materials development.