ChatGPT as a Conversational Partner in an Undergraduate English Speaking Skills Course

Abstract

Undergraduate English majors in Saudi Arabia often complete speaking courses with limited opportunities for sustained, low-anxiety oral practice. This study explored the use of ChatGPT as an out-of-class conversational partner in a Speaking Skills course taught by the researcher at Qassim University during the second semester of the 2023–2024 academic year (Spring 2024). Thirty-six male undergraduate students in two intact sections of the course took part: one section (n = 18) completed twelve structured weekly conversation sessions with ChatGPT through its mobile application, while the other section (n = 18) completed the same weekly topics through conventional pair-work practice with classmates. Speaking fluency was measured before and after the intervention through monologue and dialogue tasks analysed for speech rate, mean length of run, and pause-time ratio. Willingness to Communicate (WTC) was measured with an adapted 27-item questionnaire covering general, in-class, and out-of-class WTC. Semi-structured interviews and weekly reflective journals provided qualitative data. Consistent with the exploratory aim of the study, quantitative data were analysed with descriptive statistics only (means, standard deviations, mean gains, and percentage change), and qualitative data were analysed thematically. Both sections improved, but the ChatGPT section showed considerably larger gains: speech rate rose by 38.3% compared with 11.2%, mean length of run by 56.9% compared with 16.1%, and total WTC by 39.7% compared with 14.3%. Learners described ChatGPT as a patient, non-judgemental, and always-available partner that lowered their speaking anxiety, although they continued to value human interaction. The findings suggest that structured ChatGPT practice can be a useful supplement to classroom speaking instruction in Saudi undergraduate programmes, and they offer a practical protocol that course instructors can adapt. Because the analysis is descriptive and the sample small, the results indicate trends rather than generalisable effects.