Voices of the crowd: Exploring user evaluations of ChatGPT using structural topic modeling

ChatGPT represents a significant advancement in Human-Computer Interaction, yet public perceptions of the technology remain divided and insufficiently documented. This study analyzes large-scale YouTube comments posted between 2023 and 2025 to capture how general users evaluate ChatGPT's capabilities, limitations, and societal implications. Using a computational mixed-methods approach that integrates sentiment analysis with Structural Topic Modeling, this study reveals a clear contrast between technical definitions of system behavior and how users interpret it. Perceived bias is conceptualized less in terms of statistical characteristics and more as a matter of political or ideological significance, aligning with common folk theories about algorithmic behavior. The findings also identify a utility-anxiety tension: while users praise ChatGPT for improving efficiency in programming, structured reasoning, and creative writing, they simultaneously express concerns about factual reliability, content authenticity, and the potential displacement of both low-skill and knowledge-based jobs. Importantly, negative evaluations stem from specific encounters with error or inconsistency rather than generalized resistance to AI, whereas positive evaluations center on clear instrumental gains. The observed patterns indicate that attitudes among users are determined by particular circumstances and direct engagements, rather than by abstract assumptions. Overall, the study provides a psychologically informed account of public interpretation of generative AI, offering insights that can guide developers and policymakers in aligning system design, governance practices, and risk communication with the lived experiences and concerns of general users.