Generative Artificial Intelligence (GAI) tools are creating new opportunities to support student learning, especially for those studying mathematics in a foreign language. The present study investigates how secondary students at a private school in Venezuela, whose primary language is Spanish, engage with a multilingual GAI agent, developed using Microsoft Copilot Studio, to understand and solve math word problems written in English. Motivated by the need to prepare students for international college entrance exams such as the Scholastic Aptitude Test (SAT), this research aims to analyze students’ interactions with the GAI tutor and evaluate its effects on their persistence and confidence in mathematics. The Dynamic Strategic Math (DSM) framework and self-efficacy based on Social Cognitive Theory will be integrated to explore how GAI-supported scaffolding influences the students’ abilities in solving math word problems. A multi-case study and convergent parallel mixed-methods design will enhance the study’s rigor. By identifying the benefits and challenges of integrating AI tools into educational support systems, this study aims to provide valuable insights for developing and implementing effective, personalized AI tutors and promoting equitable access to educational resources for students from diverse backgrounds.