Performance Analysis of General and Custom GPTs in L2 Personalized Interaction

The emergence of custom GPTs has opened new prospects for various disciplines in higher education, as these models can enhance output relevance through targeted instructions and specialized knowledge bases. This study compares chat logs from both general and custom GPTs in terms of linguistic output, prompt execution, and user operation, based on data collected from a beginner-level Chinese language course. The research shows that the general GPT can only achieve appropriate vocabulary choices with an average of six rounds of interaction, whereas the custom GPT maintain effectiveness for an average of 13 rounds. Furthermore, over 80% of students expressed a clear preference for using custom GPTs due to their convenience, appropriateness in vocabulary selection, and accuracy in language feedback, citing their personalized interaction capabilities as a valuable supplement to classroom instruction. These findings provide strong evidence for the effective integration of GPTs into educational contexts, demonstrating their potential to enhance learning outcomes and support personalized educational experiences.