Large Language Model applications (LLMapps) increasingly rely on prompt templates: reusable instructions with placeholders, structure, and output constraints. Yet most templates are crafted ad hoc, which hampers reliability, maintainability, and downstream usability. This summary distills the core results from our paper published in Proceedings of the 33rd ACM International Conference on the Foundations of Software Engineering: 'From Prompts to Templates: A Systematic Prompt Template Analysis for Real-world LLMapps', a large-scale empirical study of real-world prompt templates, a component & placeholder taxonomy, and controlled tests showing which design patterns improve instruction following and structured-output stability in practice.
Yuetian Mao (Thu,) studied this question.