Language Choice and AI-Mediated Code-Switching in Tunisians’ Interactions with ChatGPT: Age-Related and Task-Domain Variation

This study examines self-reported language choice and code-switching among 150 Tunisian users of ChatGPT. A sequential explanatory mixed-methods design combined a structured questionnaire with 30 semi-structured interviews. Participants reported using Modern Standard Arabic, French, English, and mixed-language practices across task domains. A Pearson chi-square test showed an association between the four study-specific role groups and primary language choice, χ²(9, N = 150) = 29.73, p < .001, Cramér’s V = .257. A six-predictor binary logistic regression for English-primary use was not statistically significant, χ²(6) = 10.25, p = .114, Nagelkerke R² = .106; none of the individual predictors reached p < .05. Sensitivity analyses using Firth penalisation and alternative specifications of the Arabic Dominance Index supported a cautious, specification-sensitive interpretation. Eighty-one participants (54.0%) reported switching languages within a ChatGPT session; the rate was higher in the combined son/daughter groups (57/70, 81.4%) than in the combined father/mother groups (24/80, 30.0%), z = 6.30, p < .001. Interview accounts linked language choice and AI-mediated code-switching to linguistic ease, cultural and contextual appropriateness, domain-specific proficiency, task type, and perceived response quality. The findings suggest that conversational AI can operate as a perceived multilingual affordance space, while convenience sampling, self-report data, and confounding among role, age, gender, education, and occupation limit generalisation.