Despite the building up of research on the adoption of automated writing evaluation (AWE) systems, the differential effects of automated written corrective feedback (AWCF) on errors with different severity levels and gains across writing tasks remain unclear. Thus, this study fills in the vacuum by examining how AWCF through Grammarly affects college students' overall writing accuracy and errors with different severity levels. Using a quasi-experimental design, the findings demonstrate the potentials of AWCF in improving students' writing accuracy. The results were primarily attributed to the ability of AWCF to promote noticing, provide an adaptive metalinguistic explanation, and engage students in self-directed learning. Nonetheless, some challenges were reported, such as overcorrection, cognitive overload, and limited metalinguistic explanation. Implications for writing pedagogy and future studies are discussed.