AI-assisted translation has gained increasing attention in recent years, yet its effectiveness remains underexplored. The present study sought to shed light on the role of AI (ChatGPT) in mediating translation. To this end, 46 postgraduate ESP students majoring in three sub-disciplines of politics (across three classes) were selected through convenience sampling. No outliers were identified in these classes, and each was assigned to one experimental group (AI-assisted group, N = 16; AI–Human Revised group, N = 16; Human-only Scaffolded group, N = 14). A posttest-only control group design was adopted, and each group was mediated according to its respective intervention protocol. AI was instructed to follow a graduated mediation protocol developed for the purposes of this study. The final translations were evaluated both qualitatively and quantitatively. Findings revealed that the end product of the AI-assisted group, compared with the human-involved groups, exhibited major translation deficiencies ranging from the lexico-semantic level to syntax, the syntax–semantic interface, and rhetorical patterns. Additional procedural deficiencies were also observed and reported. Furthermore, participants’ translations were assessed using a rubric, and quantitative analysis showed that both human-involved groups significantly outperformed the AI-only group.