Learner perceptions of AI-mediated affective support in L2 English learning: A cross-sectional survey

Emotional barriers can shape participation in second-language (L2) learning, yet evidence about affective support attributed to artificial intelligence (AI) remains conceptually and methodologically uneven. This cross-sectional online survey examined university-affiliated L2 English learners' reported emotional challenges, perceptions of AI-mediated affective support, and views of the boundary between AI and teacher support. The Chinese-language, researcher-developed questionnaire used categorical and multiple-response items rather than psychometric scales. Of 425 submissions, one non-university respondent and four respondents younger than 18 years were excluded, yielding an adult analytical sample of 420. The non-probability convenience sample was institutionally concentrated and should not be interpreted as representative of all Chinese university L2 learners. Frequencies, percentages, and three Pearson chi-square tests were calculated. Helplessness (59.05%), fatigue (55.95%), and anxiety (53.33%) were the most frequently reported emotional challenges. Overall, 67.38% perceived that AI could provide affective support, and 84.76% reported using AI for learning challenges. Immediate feedback was the most frequently selected support mechanism (79.52%), while relief of learning pressure or anxiety was the most frequently selected function (70.71%). Perceived AI affective support was associated with reported motivation change, χ²(2) = 53.27, p < .001, Cramér's V = .356, and interest change, χ²(2) = 59.57, p < .001, V = .377. The combined belief that AI might fill or replace teacher affective support was associated with continuing need for teacher support, χ²(1) = 72.97, p < .001, φ = .417. These results indicate patterned learner perceptions, not causal effects or evidence that AI possesses genuine emotion. AI may offer scalable, low-pressure support for routine learning, whereas teachers remain important for relational, contextual, and pedagogical judgment. Longitudinal studies using validated measures are needed.