Examining user migration intention from social Q&A communities to generative AI

The research investigates the shift in user behavior from traditional social Q&A platforms (like Quora and Zhihu) to generative AI platforms (like ChatGPT). The proliferation of information and fast-paced lives drive individuals to seek efficient knowledge acquisition. Social Q&A communities offer a platform for knowledge exchange, but they suffer from challenges like low-quality user-generated content (UGC), information overload, and community fatigue. Generative AI presents a compelling alternative, offering real-time, accurate, and free answers, potentially leading to user defection from social Q&A communities. Understanding this migration is crucial for the sustainability of social Q&A platforms. While existing research explores generative AI in various fields, limited studies examine its impact on knowledge Q&A user behavior and migration patterns. This study addresses this gap by employing the push-pull-mooring (PPM) model to analyze user migration intention, utilizing a mixed-methods approach combining Structural Equation Modeling (SEM) and Fuzzy-set Qualitative Comparative Analysis (fsQCA). The PPM model allows for a comprehensive investigation of the factors driving the migration, encompassing push factors (driving users away from social Q&A), pull factors (attracting users to generative AI), and mooring factors (influencing the decision to switch).

The literature review examines user migration in cyberspace, drawing parallels between physical and online migrations. It categorizes user migration into three types: (1) migration between different media (e.g., offline to online), (2) migration between platforms offering similar services, and (3) migration within the same medium but with different characteristics. The study focuses on the third category – the migration from social Q&A communities to generative AI. The Push-Pull-Mooring (PPM) model is adopted as the theoretical framework. The PPM model, initially used for population migration studies, has been successfully applied to information systems to explain user migration. Push factors refer to negative aspects of the existing platform that drive users to seek alternatives. Pull factors represent the positive attributes of the new platform attracting users. Mooring factors are those elements that either restrain or encourage migration, such as social influence, sunk costs, continuance costs, and switching costs. This study draws upon previous research using PPM to investigate similar migration scenarios like the shift from offline to online medical services, online to offline shopping, and between various social networking sites and blogs to microblogs. The review lays the groundwork for developing hypotheses about push, pull, and mooring factors that influence user migration from social Q&A communities to generative AI.

The study employed a mixed-methods approach, combining quantitative analysis with Structural Equation Modeling (SEM) and qualitative analysis with Fuzzy-set Qualitative Comparative Analysis (fsQCA). A questionnaire was developed with items adapted from existing literature, measuring nine variables with 3-4 indicators each. A pretest was conducted with 20 users to refine the questionnaire. The questionnaire was then distributed via social media platforms such as WeChat and Weibo. The study aimed to collect data from users who had experience with both social Q&A communities and generative AI. A total of 532 questionnaires were collected, with 483 deemed valid after screening for incomplete responses, short response times, and lack of attention. The sample was predominantly young (92.8% aged 20-29), well-educated (63.8% with bachelor's degrees or higher), and largely female (55.3%). The participants utilized various social Q&A communities and generative AI platforms, reflecting a diverse user base. SEM was used to test the hypothesized relationships between variables, examining the individual effects of each variable. fsQCA was then employed to investigate the complex interplay of factors and identify different configurations that contribute to user migration intention. This approach allowed for the examination of both individual effects and the combined influence of multiple factors in determining migration intention.

The SEM analysis revealed that information overload significantly influences community fatigue, which in turn, impacts migration intention. Perceived anthropomorphism, perceived accuracy, and perceived trustworthiness of generative AI significantly influenced flow experience, which also positively impacted migration intention. Social influence emerged as a significant predictor of migration intention. The fsQCA analysis identified three main paths leading to migration intention. The first path involved information overload, community fatigue, perceived accuracy, perceived trustworthiness, flow experience, and social influence as key factors. The second path was similar but replaced community fatigue with perceived anthropomorphism. The third path emphasized community fatigue, perceived anthropomorphism, perceived accuracy, perceived trustworthiness, flow experience, and social influence, with flow experience as the core condition. SEM results showed significant effects of information overload and social influence on migration intentions. fsQCA analysis identified information overload, perceived accuracy, and flow experience as core conditions for migration intention, while community fatigue, perceived anthropomorphism, perceived trustworthiness, and social influence were identified as peripheral conditions. Low content quality did not significantly affect community fatigue according to the SEM analysis, contrasting with some existing research. The R-squared values indicated good explanatory power, with the model explaining 69% of the variance in migration intention.

The findings highlight the interplay between push, pull, and mooring factors in shaping user migration. Information overload and community fatigue are key push factors, while the perceived accuracy, trustworthiness, and anthropomorphism of generative AI, along with the flow experience it provides, are crucial pull factors. Social influence plays a pivotal role as a mooring factor, impacting users' decisions. The fsQCA results underscore the complex interactions among these factors, demonstrating that multiple combinations can lead to migration intention. The results suggest that focusing solely on individual factors may not fully capture the dynamics of user migration. The lack of significant impact from low content quality on community fatigue highlights that information quantity, rather than quality, might be the more significant driver of user fatigue in this context. The dominance of pull and mooring factors over push factors suggests the strong attractiveness of generative AI and the influence of social circles on migration decisions.

This study contributes to the understanding of user migration between social Q&A communities and generative AI. It highlights the importance of addressing information overload and user fatigue within social Q&A platforms, improving user experience through accurate and trustworthy AI, and leveraging social influence to promote generative AI adoption. Future research should consider exploring emerging generative AI models, examining other factors influencing migration intention, and analyzing actual migration behaviors to gain a more comprehensive understanding of the phenomena.

The study's limitations include focusing on a limited set of popular generative AI platforms and neglecting other potential factors (privacy risks, dissatisfaction, and habits). The study measured migration intention, not actual migration behavior, and the generalizability of the findings might be affected by the specific characteristics of the sample (predominantly young, well-educated, and from China). Future research should expand the scope of generative AI platforms investigated, incorporate additional factors into the model, and analyze actual migration behavior to enhance the generalizability and robustness of the findings.