AI literacy as contextual judgement: postgraduate researchers’ appropriateness judgements of generative AI use in a multilingual Chinese setting

Introduction The integration of generative AI (GenAI) into academic research has reshaped scholarly practices, yet how researchers-in-training develop AI literacy and make contextualised appropriateness judgements across the research lifecycle remains underexamined, particularly in multilingual contexts. Methods This explanatory sequential mixed-methods study investigated 126 postgraduate and doctoral students at a Chinese university. Questionnaire data measured AI literacy across four dimensions—understanding, applying, evaluating, and ethical awareness—and appropriateness judgements across five research stages. Semi-structured interviews with 19 participants were then analysed thematically to explain and extend the quantitative findings. Results Descriptive results revealed an uneven AI literacy profile, with ethical awareness scoring highest and understanding scoring lowest. Multiple regression analyses showed stage-specific associations between AI literacy and appropriateness judgements. Ethical awareness was positively associated with judgements in conceptualising, drafting, and transparency and honesty. Applying ability was positively associated with conceptualising, drafting, and revising/editing. Understanding was negatively associated with revising/editing, suggesting a “knowledge-trust paradox,” while evaluating showed no significant association across the five stages. Qualitative findings explained these patterns through an “assistance-then-verification” principle in data-related work, research experience as a critical moderator, task-dependent differences between coursework and publication-oriented writing, and a tiered disclosure strategy shaped by perceived desk-rejection concerns and normative ambiguity. Discussion The findings advance AI literacy theory from generic competence toward stage-sensitive, contextualised judgement. They also provide empirically grounded implications for postgraduate research training, academic integrity education, and institutional governance of responsible GenAI use in multilingual research writing contexts.