This study aimed to clarify the characteristics of individuals who prefer generative AI as a communication partner. An online survey was conducted with 98 male and 110 female participants aged 15 to 29 (M = 21.82, SD = 4.89) in Japan. First, we examined the frequency of using generative AI as a communication partner (for casual conversation, venting, or seeking advice, excluding studying or work-related tasks). The results showed that 45.7% of participants had interacted with generative AI as a communication partner in the past six months. For participants who had interacted with generative AI in the past six months, we asked whether they preferred generative AI or humans as a communication partner. Among them, 40 participants (42.1%) selected "Prefer Generative AI" or "Somewhat prefer Generative AI," 21 participants (22.1%) selected "About the same," and 34 participants (35.8%) selected "Somewhat prefer Human" or "Prefer Human." Thus, the proportion of individuals preferring generative AI as a communication partner was not negligible. We then examined whether these preferences were associated with social anxiety tendencies using the Social Interaction Anxiety Scale (SIAS-6; Peters, 2012). Analysis of variance revealed a significant main effect (F(2,92) = 3.13, p = .048, η² = 0.64), and post-hoc comparisons showed that those who preferred generative AI had significantly higher social anxiety scores than those with equal preference. These results suggest that individuals with higher social anxiety may avoid communication with humans and instead interact with generative AI. Additionally, beliefs about AI (AI having free will, AI's understanding of human minds, and trust in AI) were not related to AI preference, implying that AI preference may arise more from avoidance of human interaction than from positive evaluation of AI.