Abstract
This article examines the methodological potential of generative language models for developing philology students’ ability to analyse and interpret the poetics of the interior in literary texts. The empirical basis of the study is an initial dialogue with Google Gemini held in February 2026 and the responses to the same prompts put to Gemini, ChatGPT and Claude eight months later. The models’ responses were compared against predefined criteria, and their statements about the interior in Mukhtar Magauin’s novel Zharmak were checked against the text of the novel and a published literary review. The analysis showed that all the models were useful as a source of methodological guidance. The analytical questions they proposed are consistent with the theories of literary space developed by Yuri Lotman, Gaston Bachelard and Alexander Chudakov, and they used the conceptual vocabulary of the traditional Kazakh dwelling correctly. In the close analysis of a specific literary text, however, their reliability proved limited. Of Gemini’s 11 claims, only 2 were fully and 6 partly confirmed. ChatGPT confined itself to directing the user back to the text, while Claude drew on web sources. The novel's key spatial details, which none of the models quoted, emerged only through close reading. The study produced a three-stage model for teaching work with artificial intelligence in university philology courses and a seminar tasks that develops students’ critical analysis skills.