Recent advances in conversational artificial intelligence(AI), represented by systems such as ChatGPT, Perplexity, and Claude, have fundamentally transformed how consumers search for information by enabling natural language–based interactions. Howev...
Recent advances in conversational artificial intelligence(AI), represented by systems such as ChatGPT, Perplexity, and Claude, have fundamentally transformed how consumers search for information by enabling natural language–based interactions. However, due to the probabilistic prediction mechanisms of large language models(LLM), these systems frequently generate plausible yet incorrect responses—commonly referred to as hallucinations—which may lead consumers to erroneous judgments and decision-making. Against this backdrop, the present study examines the structural relationships among uncertainty of conversational AI information services(accuracy, integrity, and transparency), consumer information-processing styles(heuristic and systematic processing), suspicion, and information verification behavior in the context of conversational AI.
An online survey was conducted with 300 adult consumers aged 19 and above who had used conversational AI within the past six months. Descriptive statistics and one-way ANOVAs were used to assess overall variable levels and differences according to conversational AI usage characteristics. Partial least squares structural equation modeling(PLS-SEM) was then employed to analyze the structural relationships among uncertainty of conversational AI information services, information-processing styles, suspicion, and verification behavior.
The key findings are as follows. First, major variables related to conversational AI appeared at mid-to-high levels overall, with significant differences observed in information verification behavior depending on information-search purposes. Specifically, consumers who used conversational AI for educational purposes exhibited higher levels of information verification, whereas those who used it for practical convenience or leisure purposes showed relatively lower verification tendencies. No significant differences were found across usage time patterns, suggesting that the contextual purpose of use, rather than the frequency or duration of use, plays a more critical role in explaining consumers’ verification behavior.
Second, perceived uncertainty regarding accuracy and integrity significantly increased suspicion, which in turn facilitated information verification behavior. This finding highlights suspicion as a psychological trigger that encourages consumers to verify AI-generated information rather than accept it uncritically. In contrast, uncertainty related to transparency did not significantly predict suspicion or verification behavior, indicating that consumers may already perceive AI information-generation processes as relatively transparent, thereby limiting the role of transparency uncertainty in activating verification.
Third, systematic processing exerted both direct and indirect positive effects on verification behavior, confirming partial mediation through suspicion. This suggests that when consumers engage in analytical information processing, suspicion functions as an additional motivational force that promotes verification. Conversely, heuristic processing showed only a negative direct effect on verification behavior, with no significant indirect effect through suspicion. This indicates that intuitive and experience-based processing of AI responses reduces consumers’ likelihood of engaging in rigorous evaluation, as suspicion is insufficiently activated to trigger verification efforts.
This study makes several contributions. First, it foregrounds information verification—an aspect relatively overlooked in prior consumer research—within the emerging context of conversational AI. Second, by structuring uncertainty of conversational AI information services into three dimensions—accuracy, integrity, and transparency—this study refines a previously broad concept into a framework that is better suited to the AI context. Third, it examines heuristic and systematic processing as independent influences on verification behavior, acknowledging that the two modes may operate concurrently depending on usage context. Finally, it identifies suspicion as a central mediating mechanism that activates verification behavior, reframing suspicion not as a merely negative response but as a protective cognitive strategy in AI-driven information environments.