Sergio Fuentes

This study examines how five generative AI systems (ChatGPT, Claude, Gemini, DeepSeek and Qwen) respond to questionnaires commonly used in higher education to assess reading-related and emotional dimensions in human learners, with the aim of informing more sustainable classroom use. Given the high water and electricity consumption associated with intensive GenAI use in universities, optimising tool selection based on response effectiveness is increasingly necessary. Using a descriptive quantitative approach, the Metacognitive Awareness Inventory (MAI) and the Wong–Law Emotional Intelligence Scale (WLEIS) were applied as human-oriented questionnaires. The analysis focuses on the degree of alignment between AI-generated answers and predefined reference response patterns. Results show clear differences among systems. Gemini and ChatGPT exhibited comparatively stable alignment in reading-related items, particularly in attention and memory, while Qwen showed the lowest overall deviation in both reading-related and emotional self-report items. Claude and DeepSeek displayed greater variability across both domains. These findings suggest that response consistency and task-specific alignment may serve as practical criteria for selecting generative AI tools in higher education, supporting pedagogical effectiveness while contributing to more environmentally responsible digital practices aligned with the Sustainable Development Goals.