AI-Powered K-12 Instruction: Educator-AI Interaction, Teacher-Authored Student-AI Interaction, and Nudges

Abstract

This dissertation positions educator agency as a core condition for making generative AI (GenAI) pedagogically meaningful and classroom-ready. Across educator-facing and classroom-facing contexts, the central problem is not whether AI can generate fluent content, but whether educators can reliably translate pedagogical intent into usable artifacts and productive interactions under authentic constraints such as time pressure, incomplete information, competing priorities, and the need to balance rigor, equity, and feasibility. The dissertation synthesizes three complementary studies. Study 1 analyzes large-scale authentic educator-AI conversations and characterizes educators' AI-assisted work as iterative professional design, marked by repeated cycles of ideation, specification, repair, validation, and formatting to reach classroom-ready products. Study 2 investigates the classroom deployment of a teacher-configured student-AI conversational feature, showing how teacher-authored prompts function as scaffolds that transmit and preserve their pedagogical intent in student-AI interactions. Study 3 examines pedagogically grounded, AI-generated next-step nudges embedded in educator-AI conversations across a large-scale deployment, showing that adoption is selective yet widespread and increases when nudges are immediately executable and effort-reducing. The findings show that classroom-ready GenAI depends on tool designs that preserve teachers' professional expertise and pedagogical intent while reducing workflow friction and providing practical, evidence-grounded decision support that helps educators translate intent into action.