Can Generative <scp>AI</scp> Empower Tibetan Students in English Writing? Exploring Performance Gains, Self‐Efficacy Enhancement and Cultural Expression

ABSTRACT Background Tibetan university students often write in English within a complex multilingual environment involving Tibetan, Chinese and English. Previous studies have suggested that multilingual learners may experience challenges related to cross‐linguistic transfer, as well as difficulties in expressing culturally specific concepts in English writing. At the same time, limited research has examined how contextually relevant AI‐supported feedback is experienced in minority‐language educational contexts. Objective This study explored associations between participation in a contextually relevant GenAI‐supported writing intervention (implemented through the Doubao app) and Tibetan college students' English writing performance, self‐efficacy (SE) and experiences of cultural expression in English writing. Methods A 16‐week quasi‐experiment was conducted with 60 Tibetan students paired into an experimental group (EG, n = 30, using AI with cultural support) and a control group (CG, n = 30, traditional teaching). Data were collected through writing tests, an SE scale and semi‐structured interviews. Results The EG demonstrated higher scores than the CG on several writing measures, particularly Vocabulary and Language Use, and these differences remained evident in the delayed post‐test. Positive patterns were also observed across writing SE measures. Interview data suggested that some participants perceived cultural terminology support and contrastive linguistic feedback as helpful during writing and revision activities. Conclusion The findings suggest that participation in a contextually relevant AI‐supported writing intervention was associated with positive writing outcomes among Tibetan trilingual learners. Participants perceived feedback addressing Tibetan–English structural differences and culturally specific terminology as helpful during writing tasks. Further research is needed to better understand the specific contribution of these context‐sensitive features.