UC Riverside

This study examines how AI influences existing differences in professional writing between native and non-native English writers (NEWs and NNEWs) in the United States, reflecting individuals’ task-related proficiency and language-based social positioning. We compare how these two groups integrate AI-generated content into their writing and include writing self-efficacy as a moderator to examine whether perceived task proficiency shapes the differences in AI use between the two groups. We also test an underlying social-psychological mechanism by examining the mediating role of perceived superiority: the extent to which participants judged the AI-generated text as better than their own writing. In an online experiment, 327 NEWs and NNEWs were recruited to write a job application cover letter for a hypothetical scenario. Participants were randomly assigned to receive AI-generated content written at either a simple or advanced lexical level and were asked to revise their letter as they chose, potentially incorporating the AI-generated content. Using natural language processing techniques, we measured the extent to which participants integrated AI-generated content. Findings show that NNEWs favored simpler content, while NEWs tended to incorporate advanced content. This difference was pronounced among those with lower writing self-efficacy. However, the perceived superiority of AI-generated text did not explain this pattern. These findings show that while AI can support lower-skilled groups in specific tasks, sociolinguistic gaps still remain even with AI-assistance. We conclude by calling for AI training frameworks that scaffold learning to address both proficiency gaps and social constraints.