Information Disclosure as a Behavioral Manifestation of Trust in Anthropomorphic LLM-Based Conversational Agents: Investigating the Role of Perceived Trustworthiness and Privacy Concerns from an Integrative Model of Trust Perspective

Large language model-based conversational agents (LLM-CAs), such as ChatGPT, have been growing exponentially in popularity over the past few years and have quickly become a part of people’s everyday lives (Lalot & Bertram, 2025; Motoki et al., 2024). Despite the usefulness of this technology, users report concerns about a lack of transparency regarding the treatment of their data when disclosing personal information to AI (Choung et al., 2023; Menard & Bott, 2025). Moreover, LLM-CAs, such as ChatGPT, are designed to learn from interactions with users, personalizing their responses based on information provided by the user (Kelly et al., 2025). Thus, users’ disclosure of personal information is closely related to their usage experience, while potentially posing a privacy risk to them, as recent reports on the misuse of non-disclosure agreements by the AI developer OpenAI demonstrate (Milmo, 2024). Research regarding the relationship between privacy concerns and disclosure of personal information in the context of LLM-CAs is still in its infancy due to the relative newness of this technology. Nevertheless, there is evidence suggesting that users with greater privacy concerns are also less willing to disclose personal information to LLM-CAs (Gumusel, 2025; Ischen et al., 2020; Lappemann et al., 2023). Furthermore, researchers have identified trust as a key determinant of information disclosure to LLM-CAs (Hoogerheijde, 2024; Lalot & Bertram, 2025; Luo et al., 2023). Lalot and Bertram (2025) provided evidence for the applicability of the Integrative Model of Organizational Trust by Mayer et al. (1995) to an LLM-CA context, discussing the anthropomorphic nature of LLM-CAs. Moreover, researchers in e-commerce have identified perceived trustworthiness as a statistical mediator between privacy concerns and willingness to disclose personal information (Malhotra et al., 2004; Metzger, 2004). Although the Integrative Model of Trust has been widely used as a theoretical foundation to investigate trust in the past, most studies have either measured trust unidimensionally or have failed to distinguish between trustworthiness and trust as willingness to be vulnerable (Gillespie, 2011; Lalot & Bertram, 2025; McKnight et al., 2002). Therefore, to date, trust-related mechanisms through which privacy concerns affect the willingness to disclose personal information to an LLM-CA remain unclear. Drawing on the conceptualization of trust provided by Mayer et al. (1995), the goal of the present study is to complement existing literature regarding the relationship between privacy and trust by investigating privacy concerns toward an LLM-CA as a statistical predictor of perceived trustworthiness of the LLM-CA and willingness to disclose personal information to it. It is hypothesized that the effect of privacy concerns on willingness to disclose personal information is statistically mediated by perceived trustworthiness.