Considering the individual differences in previous content knowledge, skill, and attitude, which types of learners are suitable for AR is a valuable but complex question. The study used quasi-experimental design, and divided 97 10th-grade students into two groups: traditional group (N = 48) and AR group (N = 49), who participated in 4-week organic microstructure teaching. Pre-test, post-test, delayed test, and remedy test were used to collect data of students' individual differences (foundation of learning in chemistry, spatial ability, and attitude towards AR) and learning (immediate and lasting) outcomes. Whether AR is used or not and three individual differences were taken as causal conditions, which jointly influence the learning outcomes. FsQCA was used to deal with the non-linear and asymmetric relationship among the causal conditions, and to obtain the configurations of good or poor learning outcomes. The results show that, (i) using AR usually contributes to a sufficient configuration of good learning outcomes, (ii) especially is beneficial to lasting learning outcomes, (iii) but it is still not the necessary condition, and (iv) using AR by some types of students causes poor outcomes instead. Which types of students are suitable or not suitable to use AR is also be discussed. The study emphasizes the importance of personalized use of educational technology.