Natural Language as the Bridge Between Pedagogy and AI

This paper presents a lightweight framework for encoding pedagogical expertise as structured constraints for large language model content generation. The framework decomposes educational content into discrete elements, each guided by natural language parameters that specify both per-element constraints and inter-element relationships. The paper argues that teachers already express their expertise in rule-like terms that map naturally onto the structured constraints known to improve LLM output quality, and that this convergence creates a practical opportunity for educators to produce pedagogically coherent content at scale without technical knowledge. The framework is demonstrated through worked examples ranging from simple three-element pedagogies to a six-element Rosenshine-style lesson plan, with copy-paste prompts that readers can use immediately in any LLM. The paper also describes batching strategies for generating entire content libraries from a single pedagogy, approaches for inferring pedagogies from existing materials, and the conditions the framework creates for systematic pedagogical comparison. This is the first of four papers; subsequent papers extend the architecture to audience tailoring, visual design, and temporal sequencing for video production.