This paper analyzes the use of speech levels and shifts between Chinese students of Japanese and artificial intelligence (AI) dialogue systems in compositions written in Japanese. The results indicate significant differences in terms of speech levels. The AI emphasized stylistic consistency and showed a preference for the use of the polite form of words, while some of the students did not distinguish between the polite and non-polite forms, which led to confusion in terms of speech-level usage. For instance, employ the non-polite form for teachers and the polite form for friends and other peers. Conversely, although both the students and the AI engaged in shifts, the errors made by the students were more complex. The factors that contributed to these differences primarily involved the distinct mechanisms of Japanese learning, emotional output, and cognition.