This study presents a comparative analysis of ChatGPT 4.0 (see further ChatGPT) and Gemini AI 2.0 (see further Gemini AI) models, evaluating their efficiency in adapting reading materials for undergraduate business students. In the rapidly evolving landscape of education, integrating artificial intelligence (AI) into learning materials design has become a critical field of investigation. Educational resources often require tailoring to accommodate diverse learners’ needs and backgrounds as their accessibility plays an important role in undergraduate students’ academic success. Manually adapting texts can be time-consuming and resource intensive. AI language models like ChatGPT and Gemini offer a possibility to automate this process, potentially alleviating pressure on educators and increasing content customization opportunities.

The aim of the current study is to identify the AI platform that both excels in content quality and simplifies the workflow, optimizing teaching material design for ESP purposes. To reach this aim, a method of comprehensive comparative and contrastive analysis of ChatGPT and Google Gemini models was conducted, and their capabilities in adapting textual content were examined. The models were evaluated based on several criteria: reliability for reading ease evaluation, accuracy in reading manipulation on specific parameters, and prospects of improving user experience. There were several prompts designed to investigate efficiency of the AI models for text adaptation. Multiple content regeneration attempts were taken to explore consistency of the AI output. The study considered text, sentence, and word length, vocabulary choice and content preservation as the main features for evaluation. Flesch-Kincaid reading ease metric, Lextutor vocabulary profiler, and human assessment were the output quality measuring and comparison tools. The text complexity evaluations of the original, human-modified, and AI-modified texts were compared to identify the accuracy and effectiveness of the latter.

The results of the comparison identified the areas of concern of using Chat GPT and Gemini AI models for text manipulation and requirements for their successful implementation in the process of the reading materials transformation. Among those concerns are that (1) LLMs under study (ChatGPT 4.0, Gemini 2.0) exhibit limitations in accuracy and reliability for complex text analysis tasks, particularly readability scoring and lexical complexity assessment., (2) inconsistencies in performance across parameters (e.g., passive voice detection) and word tiers (common vs. less frequent vocabulary), (3) challenges in meeting user requirements in increasing overall text readability. The research also determined beneficial ways of AI implementation in educational material design, such as promising text synthesis capabilities of Gemini AI, as well as bringing the text within the target difficulty range. Overall, it can be concluded that LLMs with strong text simplification capabilities can be valuable in education and accessibility applications with human oversight and verification. Exploration of hybrid approaches combining the strengths of various LLMs for enhanced text analysis and modification capabilities could optimize the workflow of a teaching materials designer.

Keywords: ESP, AI, ChatGPT, Google Gemini, text adaptation.