Prompt Engineering: Techniques, Failures, and Evaluations—A Comprehensive Survey

Large language models (LLMs) have revolutionized natural language processing, with prompt engineering emerging as a critical factor in their effectiveness.This literature survey provides a comprehensive overview of current research on prompt engineering and failure modes in LLMs, systematically categorizing and organizing existing work in this rapidly evolving field.We explore the landscape of currently available prompting techniques, examining their design principles, applications, and effectiveness across diverse tasks.The survey organizes existing research into a systematic taxonomy of prompt failures, including content-related, structural, behavioral, and consistency issues as identified across multiple studies.We compile and classify assessment frameworks reported in the literature, encompassing automated metrics (BLEU, ROUGE, BERT Score), human evaluation protocols, and multidimensional quality measures.The survey maps optimization approaches documented across recent research, including genetic algorithms, gradient-based methods, and reinforcement learning techniques, illustrating the field's progression from empirical practices toward systematic methodologies.By comprehensively organizing and classifying the current body of research, this survey serves as a reference guide for researchers and practitioners seeking to understand the landscape of prompt engineering, failure modes, and optimization strategies in LLM applications.