Learning Requirements Engineering with Large Language Models: An Empirical Study on Functional Requirements Quality

Large Language Models (LLMs) have been increasingly adopted in Software Engineering education; however, empirical evidence on how structured educational interventions contribute to learning software requirements quality remains limited. This paper presents an exploratory empirical study involving 27 undergraduate students who participated in a Requirements Engineering educational intervention. Students attended a workshop on software requirements quality and Prompt Engineering before using Google Gemini 2.5 Pro to support the specification of functional requirements for seven software projects. The resulting artifacts were independently evaluated by three Requirements Engineering specialists based on seven established quality attributes, while students' perceptions were collected through a post-activity questionnaire. The results indicate that LLMs supported the initial development of functional requirements and encouraged iterative prompt refinement. The specialists' evaluations revealed satisfactory overall quality, with the highest scores observed for correctness, feasibility, and consistency, whereas completeness and verifiability remained the main challenges. The findings also suggest that integrating LLMs into Requirements Engineering education fosters competencies in Prompt Engineering, critical evaluation of AI-generated artifacts, and understanding of software requirements quality. This study contributes empirical evidence on the educational use of LLMs and discusses implications for their integration into Software Engineering education.