Prompt-as-Lesson: Authoring AI-Delivered Lessons for Fidelity and Observability

Abstract

This paper reports the development of the prompt-as-lesson: a lesson encoded as a single prompt and run in the student’s own LLM tool, with the model cast throughout as a tutor delivering the teacher’s lesson, and the session closing with a report of the dialogue returned to the teacher. The purpose of the prompt-as-lesson is to enable students to pursue teacher-specified learning outcomes through tutorial dialogue with an LLM. The interactive, conversational nature of an LLM makes Laurillard’s Conversational Framework the natural interpretive lens. Under this lens, the paper argues, the LLM is the first single medium capable of enacting the iterative teacher–learner dialogue the framework holds essential to formal teaching, yet this dialogue is frequently lacking in higher education. Introducing LLM activities into the formal learning environment risks inconsistency as, left to its defaults, the model teaches from its own conception and constructed learning environment rather than the teacher’s. The prompt-as-lesson is a response. Utilising the prompt-as-lesson within the teacher’s formal environment makes three demands: consistency, so that what is taught is drawn from the teacher’s conception and pedagogy; adherence, so that the enacted lesson remains the authored one; and reporting, so that the dialogue is not opaque to the teacher. Design-based iteration on two contrasting modules to address these demands yielded two primary findings: contemporary LLMs struggled to adopt the learner’s perspective and would not reliably hold to an unstructured lesson. The format that emerged seeks to address both issues and comprises three artefacts: a control surface of fourteen dimensions; an eight-section structure; and a four-phase authoring method. The paper claims no effect on learning; its contribution is a control surface, a structure and an authoring method. A formal evaluation of the format’s fidelity of implementation begins in September 2026.