Automated writing evaluation (AWE) systems have been increasingly incorporated in L2 writing instruction. However, how the instructional use of AWE influences learners' writing has remained inconclusive partly due to variations in heterogeneous learners. Underpinned by a sociocultural theory, this classroom-based study was to understand how English learners of different proficiency levels differed in their written products measured by complexity, accuracy and fluency, multiple revisions, and the perceived usefulness after 15-week instructional use of AWE. The results showed that disparity in writing accuracy of learners with different proficiency levels greatly reduced after the mandatory use of AWE feedback, but syntactic complexity and fluency remained unchanged. Meanwhile, perceived usefulness of AWE feedback was contingent on language proficiency. Limitations and pedagogical implications were also discussed.