AI챗봇에 대한 인지된 개인화가 고객시민행동에 미치는 영향:인지된 공감과 프라이버시 침해우려의 조절된 매개효과 = The Impact of Perceived Personalization of AI Chatbot on Customer Citizenship Behavior: The Moderated-Mediation Effect of Perceived Empathy and Privacy Concern

다국어 초록 (Multilingual Abstract)

Purpose: AI and big data have transformed customer behaviors and marketing strategies, with businesses using customer data for personalized services that improve efficiency and reduce workload. For customers, AI enhances convenience by automating tasks. Post-COVID, AI adoption accelerated, with 42% of businesses using chatbots that leverage natural language processing and machine learning. In banking, 51.4% of representatives acknowledge chatbots’ effectiveness, while retail implementation has increased 31% since 2021, now reaching 73% of global companies. Current research limitations include focus on functional rather than intelligent features, limited empirical research on personalization and psychological responses, insufficient research on empathy’s influence, minimal privacy concern research, and poor understanding of interaction factors. This study examines chatbot personalization’s impact on perceived empathy, how empathy affects customer citizenship behavior, and privacy concerns’ moderating effect. The research draws on CASA theory, AI Job Replacement Theory, AI Device Use Adoption model, and Information Boundary Theory. Based on this theoretical background, the study developed the following hypotheses: [H1] Perceived personalization of AI chatbot services will increase customer citizenship behavior.
[H2] Perceived empathy will mediate the effect of perceived personalization of AI chatbot services on customer citizenship behavior.
[H3] Privacy concerns will moderate the mediating pathway where perceived personalization of AI chatbot services affects customer citizenship behavior through perceived empathy.
Research design, data and methodology: This study surveyed Korean adults who had used AI chatbots, following Hair et al.’s (2010) guidelines recommending 5–15 responses per observed variable for structural equation modeling. With 14 variables, they targeted 140–210 responses. The online survey ran February 1–15, 2024, through a domestic panel agency, screening for those who had used AI chatbot customer service within the past year. From 234 participants, 24 were excluded for short response times or irrelevant answers, leaving 210 respondents. The sample was 64% male with an average age of 32.1 years, distributed across age groups: 30s (39%), 20s (31%), 40s (23%), and 50s (7%). For chatbot service frequency, 41.1% reported “less than 10 times,” 21.9% “20+ times,” 19.4% “less than 5 times,” 10.6% “less than 15 times,” and 7.0% “less than 20 times.” With 81.6% having used chatbots at least 5 times, the sample confirmed their widespread use. Measurement items used a 5-point scale, with perceived personalization measured by four items, perceived empathy by three items, customer citizenship behavior by four items, and privacy concerns by three items.
Results: Using Hayes’ PROCESS MACRO Model 4, the total effect of perceived personalization on perceived empathy was significant (b=0.671, p<0.001), supporting Hypothesis 1. Bootstrapping analysis (n=1,000) showed a significant indirect effect of perceived empathy (0.273, 95% CI=[0.186, 0.371]) and significant direct effect (95% CI=[0.282, 0.514]), confirming perceived empathy partially mediates the influence of perceived personalization on customer citizenship behavior, supporting Hypothesis 2 (show <Table 1>).
To verify whether privacy concerns moderate the indirect effect of perceived personalization of AI chatbot services on customer citizenship behavior through perceived empathy, Hayes' PROCESS MACRO Model 7 was used. The analysis showed a statistically significant interaction effect between perceived personalization and privacy concerns on perceived empathy (b=–0.160, p<0.001). This confirms that the influence of perceived personalization on perceived empathy varies depending on the level of privacy concerns (see <Table 2>).
A conditional effect analysis examined the moderating effect of privacy concerns, which were classi...