Grammarly is one of the famous Automated Writing Evaluation (AWE) tools used by EFL students. Many studies have revealed that this tool can improve students’ writing. However, studies evaluating Grammarly’s feedback accuracy, especially the premium version, are still limited. This study aims to: (1) explore EFL students' perceptions of Grammarly premium feedback, (2) provide a detailed analysis of types of inaccurate feedback provided by Grammarly premium, and (3) investigate how students handle inaccurate feedback. This mix method study used a questionnaire, semi-structured interviews, and documentation to collect the data. The participants were undergraduate EFL students who were writing or had completed their thesis. The results indicate that: (1) students have positive and negative perceptions on the use of Grammarly Premium. They claimed that Grammarly Premium helped them in writing. The ‘correctness’ feature improved spelling, grammar, punctuation, convention, article, and consistency. The ‘clarity’ feature improves readability, such as conciseness and passive voice. The ‘engagement’ feature made writing varied and the ‘delivery’ feature to get the right tone. On the other hand, students complained about the need of a stable internet connection to work properly and require additional proofreading due to some inaccuracies. (2) Inaccurate feedback generally occurred in the features of correctness (convention, spelling, punctuation, article, and consistency), clarity (passive voice and conciseness), and engagement (variety). (3) Students continued to use Grammarly Premium because it helped them self-proofreading even though it had some shortcomings. In short, students still preferred to use Grammarly Premium with some concern in accepting the provided feedback.