Keeping Learners in the Optimal Challenge Zone: Generative AI–Based Task Personalization in College EFL Classes

ABSTRACT In college English as a Foreign Language (EFL) classes, a mismatch between task complexity and learners’ speaking skills can reduce engagement and limit learning outcomes. To address the difficulty of implementing differentiated task grading in large‐scale classrooms, this study introduces generative AI (GenAI) as the control center and examines the personalized task complexity regulation mechanism based on learners’ immediate performance and emotional feedback. The study adopted a 10‐week quasi‐experimental design. A total of 70 first‐year undergraduate students from a university in western China participated in the study, with 35 students assigned to the experimental group and 35 to the control group. Both groups completed multiple rounds of speaking tasks under the same conditions and with the same communication goals. After each round, the experimental group collected data on complexity, accuracy, and fluency (CAF), self‐assessed task difficulty, and anxiety, and adjusted the complexity of subsequent tasks accordingly. The control group increased the difficulty in a preset uniform sequence. The results showed that personalized task complexity adjustment significantly reduced the proportion of mismatch between tasks and skills, increased the proportion of strong match, and improved learners’ CAF performance. Mechanism analysis showed that incorporating difficulty self‐evaluation and anxiety into adjustment decisions helped avoid inefficient investment caused by overly difficult or overly easy tasks while enhancing the stability and adaptability of learners’ speaking development. This study proposes and validates a classroom‐implementable GenAI‐driven dynamic regulation path of task complexity that integrates cognitive and emotional cues, providing an empirical basis for the design of differentiated EFL speaking tasks.