Designing effective generative AI support requires understanding how people choose and use these tools for their information tasks. We conducted an exploratory survey study examining what motivates people to use ChatGPT for information tasks, how they perceive its response quality across tasks of varying familiarity, and what concerns arise during use. We collected 110 survey responses from Amazon Mechanical Turk participants about two recent ChatGPT-supported tasks (220 tasks total), including motivations for use, evaluations of responses (specificity, relevance, accuracy, understandability, creativity), and open-ended reflections on overall experience. Qualitative coding revealed 12 motivations spanning technology capabilities, task demands, and individual preferences, while perceived fit centered on ChatGPT’s capabilities, response quality, efficiency, and support for non-routine tasks. The tasks participants identified as relatively easy tended to be short, low-stakes, and well-understood, and participants rated ChatGPT’s responses for these tasks higher on four of five evaluation dimensions. Tasks participants identified as more challenging tended to involve lower self-reported familiarity, more requirements or constraints, and lower perceived response quality on specificity, relevance, accuracy, and understandability, while creativity ratings did not differ reliably. Across tasks, concerns about accuracy, technology limitations, and user AI literacy often led to verification or rework. Taken together, the findings provide a descriptive account of how users’ motivations, perceived response quality, and retrospective assessments of fit vary across familiar and unfamiliar information tasks, and point to designs that better adapt capabilities and verification support to task demands.
Li et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: