This study examines the conceptual nature of prompts as the point of departure for image generation in generative AI environments and explores their educational significance in the context of elementary art education. The rapid expansion of generative AI raises fundamental questions about long-standing assumptions in art education concerning expression, creation, and assessment. In particular, prompts occupy a contested position: they may be understood as technical inputs, expressive acts, or acts of selection. To address this issue, the study conducts a literature-based inquiry and conceptual analysis of generative AI, artistic creation, visual culture, and art education theory. The analysis suggests that prompts function as a form of dual language situated between natural language and algorithmic operation. Rather than determining a resulting image in a one-to-one manner, prompts organize a probabilistic field in which images may emerge. They also possess both an expressive dimension, through which users verbalize visual intentions, and a selective dimension, through which the range of generable images is narrowed. These two dimensions do not operate separately. Accordingly, prompts should be reconceptualized not as direct acts of making, simple technical commands, or mechanical choices, but as mediating discursive acts in which expression and selection intersect. This reconceptualization implies the need to introduce prompt literacy, design mixed-media learning experiences connected to material engagement, specify process-oriented assessment criteria, and strengthen teachers' interpretive professionalism in elementary art education.