Pre-service Teachers Comparative Evaluation of Large Language Mod-els: A survey among Learning Sciences

The rise of generative Artificial Intelligence (AI) necessitates moving beyond digital competencies to AI literacy for educators. This study examines preservice primary teachers' knowledge, perceptions, and preferences regarding generative AI tools, specifically comparing ChatGPT and Gemini for instructional design tasks. Understanding user perceptions can help tailoring effective teacher training pathways. A sample of 172 pre-service teachers completed three comparative tasks using both ChatGPT and Gemini: gen-erating lesson designs from given prompts, creating prompts following guidelines, and gen-erating mathematical problems. Participants rated models on six performance dimensions and provided open-ended justifications. Data were analyzed using qualitative content analy-sis of 721 responses and segmentation analysis across experience levels, gender, and task types. ChatGPT received majority preference (67% overall), with consistent superiority across all tasks and experience levels. Qualitative analysis revealed three primary evaluation criteria: completeness, coherence, and organization. Experienced users showed stronger ChatGPT preference (72%) compared to novices (61%). Task complexity moderated pref-erences, with ChatGPT demonstrating strongest advantages in creative and analytical tasks. In conclusion, User preferences are driven by functional attributes—particularly response organization and coherence—rather than advanced features. Effective teacher training re-quires differentiated pathways based on prior experience, explicit instruction in prompt en-gineering, and continuous assessment of evolving AI tools.