WER2026 - 29th Workshop on Requirements Engineering


Evaluating Generative Artificial Intelligence for Teaching User Stories

Claudia Marcos; Guillermo Rodríguez; Delfina Villarreal Garrido

10.29327/1884517.29-7

PDF Scholar

Abstract

Generative Artificial Intelligence (GAI) has demonstrated significant potential as a pedagogical assistant across various disciplines. In Software Engineering, User Stories (US) constitute a fundamental technique for requirement specification from the user's perspective, relying on the "Three Cs" framework (Card, Conversation, and Confirmation) to ensure functional and non-functional understanding. Nevertheless, teaching this technique in massive university environments poses a significant challenge due to the requirement for continuous feedback. This paper presents a study structured into four experiments with incremental context levels to evaluate GAI's capability to generate US aligned with specific pedagogical objectives. The methodology ranged from the exclusive provision of problem statements to the inclusion of detailed theoretical material and faculty-resolved examples. The results, evaluated through direct comparison with official solutions and compliance with the INVEST criteria, demonstrate that model precision significantly improves as domain-specific information is provided. The scenario integrating both theory and practice achieved the highest performance. However, persistent limitations were detected regarding the models' ability to adjust granularity and establish coherent relationships between stories. It is concluded that GAI is a valuable tool for fostering autonomous learning, although its use does not replace the supervision of a specialized instructor.

Keywords: Generative Artificial Intelligence; User Stories; Software Engineering Education; INVEST criteria; Pedagogical Assistance