Generative AI Use and Authorial Voice among Pakistani University ESL Learners

Abstract

Generative artificial intelligence (GenAI) is rapidly transforming second-language academic writing, but its association with authorship and authorial voice remains insufficiently understood. This explanatory sequential mixed-methods study examined whether editorial use, dialogic use, and generative delegation were differentially associated with authorial voice among 92 university ESL learners in Southern Punjab, Pakistan. Quantitative evidence comprised survey measures, one argumentative essay per participant, corpus-based linguistic indicators, and independent ratings using an analytic voice rubric; qualitative explanation was provided by interviews with 14 purposively selected participants. Generative delegation was negatively correlated with self-perceived authorial voice (r = -.45, p < .001) and rubric-scored textual voice (r = -.44, p < .001), whereas dialogic use was positively correlated with self-perceived voice and editorial use was not significantly related to either voice outcome. The final regression model accounted for 52.0% of the variance in self-perceived voice; writer agency had the largest standardized coefficient. The reported mediation analysis indicated that agency accounted for 43.6% of the association between delegation and self-perceived voice, whereas the delegation-by-AI-literacy interaction was not statistically significant. Greater delegation was also associated with fewer self-mentions and hedges, but with more boosters, longer sentences, and greater lexical diversity. Interviews generated four themes: outsourcing the difficult middle, the fluency-ownership trade-off, the disappearing draft, and voice as risk management. They nevertheless suggest that pedagogical guidance should distinguish assistance that supports writers' decisions from delegation that replaces core composing decisions.