This study examined the use of avatars to provide feedback to influence action identities, anxiety, mood, and performance during job interview training. We recruited 36 university students for the experiment and divided them into two groups. The first group received avatar-based feedback whereas the other group received self-feedback after the first interview session. Results showed that the avatar-based feedback group experienced significantly higher levels of action identities, reduced anxiety, and happier mood in the second interview after the feedback session. Additionally, compared to the self-feedback group, the avatar-based feedback group rated their performance better in the second interview. Furthermore, the effect of avatar feedback on mood and performance varied depending on the participants' initial anxiety levels. For those with low initial anxiety, avatar feedback resulted in higher levels of action identification, a marginally significant increase in positive mood, higher pitch, better word usage, and better self-assessment. In contrast, for those with high initial anxiety, avatar feedback did not cause any significant changes in the action identities or moods, however, it reduced anxiety and the use of weak words. We believe that our findings have implications not only for improving job interview performance but also for designing future communication coaching systems.