This study aims to perform an emotional analysis of self-expressive writing by student authors using large language models. We collected two pieces of writing for each of the most- and least-preferred emotion words selected by 169 university students from Russell’s emotion-term list for a total of 338 texts. Using the most accurate method—ChatGPT with an emotion dictionary as a prompt—we compared the frequencies of posi- tive and negative words used. For the preferred emotions, we examined negative word characteristics and explored the potential of positive psy- chological therapy. For the least-preferred emotions, we confirmed that the distribution patterns of emotion words varied according to the emo- tion type.