KISDI Study Finds Reliability Concerns and Errors Drive Generative AI User Attrition, While Human-Like Interaction Encourages Continued Use (detail)

※ URL(Korean): https://www.kisdi.re.kr/bbs/view.do?bbsSn=114957&key=m2101113055776&pageIndex=1&sc=&sw=

KISDI Basic Research (25-12): An Exploratory Study on the Antecedents of Generative AI Service Adoption

▲ User experience has expanded, but sustained use remains limited due to perceived gaps between expectations and practical value
▲ Verification costs caused by errors and hallucinations emerge as a major reason for discontinuing use
▲ Anthropomorphic interactions positively influence continued use intentions
▲ Reliability concerns are a decisive factor in attrition among professional users

A new study by the Korea Information Society Development Institute (KISDI, President Sangkyu Rhee) suggests that improving the reliability of generative AI services is critical for moving beyond initial curiosity and supporting sustained user adoption. The study finds that while interest in generative AI continues to grow, concerns about reliability and information accuracy remain key barriers to continued use.

KISDI recently published *An Exploratory Study on the Antecedents of Generative AI Service Adoption* (KISDI Basic Research 25-12), which examines the factors influencing continued use and user attrition in generative AI services.

The report notes that generative AI has spread rapidly due to its ability to create diverse forms of content, while related policies have largely focused on promoting technological development and adoption. However, actual usage patterns indicate that regular and sustained use remains concentrated among specific age groups and occupational categories.

To better understand user behavior after initial adoption, the study examined the post-adoption stage of generative AI use based on a representative sample covering a broad range of age groups. The research combined analysis of public discourse on YouTube with surveys of users and experts to identify factors influencing continued use or discontinuation.

The findings indicate that continued use of generative AI services depends not only on accessibility but also on users’ perceptions of trustworthiness, usefulness, and interaction quality. In particular, the study found that the need to verify information generated by AI systems due to errors and hallucinations creates additional costs for users, reducing perceived usefulness and contributing significantly to service abandonment.

At the same time, experiences involving personalized responses, conversational interactions that reflect contextual understanding, and human-like engagement were found to have a positive influence on users’ intentions to continue using generative AI services. The study also identified cases in which users engaged with AI systems in relational ways, such as discussing personal concerns or perceiving AI as a safer communication partner because it poses less risk of betrayal than human relationships. In addition, differences in digital literacy levels were found to influence users’ evaluations of and attitudes toward generative AI services.

Sung-Hee JOO, Fellow at KISDI, stated, “There is a need for user-centered policy design based on users’ experiences and perceptions. As users simultaneously recognize both the benefits and potential risks of generative AI, creating a trustworthy AI environment is becoming increasingly important.”

The report suggests that policy approaches for AI diffusion should shift from a focus on technology deployment to one centered on managing user experience. Specifically, it recommends strengthening transparency and hallucination management to reduce user attrition, providing differentiated digital literacy education to help users develop realistic expectations, and establishing user-centered governance frameworks that address concerns and anxieties related to AI technologies.

The study also emphasizes the importance of building institutional foundations that can alleviate social concerns about generative AI, including the establishment of clear organizational guidelines and support systems for AI use.

The report is available for download from the KISDI website.

KISDI News

KISDI Study Finds Reliability Concerns and Errors Drive Generative AI User Attrition, While Human-Like Interaction Encourages Continued Use