<![CDATA[Generative Artificial Intelligence (AI) tools are increasingly used in instructional design (ID) contexts; however, little empirical work has examined whether frontier language models produce theory-grounded instruction without explicit pedagogical guidance. This study compared zero-shot and theory-guided prompting across three large language models to evaluate whether principled instructional structure emerges on its own or requires explicit theoretical direction. Using a primarily qualitative instrumental case study design, each model generated a training curriculum under both conditions. Outputs were evaluated using a five-dimensional instrument grounded in Merrill's First Principles of Instruction. Theory-guided prompting produced higher overall instructional integrity than zero-shot prompting for two of the three models, with the largest gains in application and in integration and transfer. Zero-shot outputs were generally well-organized but more often reflected topic-based information presentation than principled ID. These findings suggest that, in the cases examined, even frontier-level models do not reliably produce theory-grounded instructional structure without explicit guidance, particularly in ways consistent with cognitive principles, and that generative AI cannot substitute for sound ID practice or human oversight. For instructional designers and educators, the results underscore the importance of instructional theory in AI-assisted curriculum development.]]>
Copyrights © 2027
