Laypersons’ perceptions and misconceptions of AI-chatbots: a systematic review

Recent generative AI advances have turned chatbots with pre-programmed responses into systems that perform text generation, summarization or translation and mimic human conversation. Understanding how non-experts perceive these chatbots is essential, given their widespread use, complexity, and susceptibility to errors. This paper systematically reviews empirical studies on laypersons’ perceptions and (mis-)conceptions of AI-chatbots. Following PRISMA guidelines, we conducted advanced searches in Scopus, Web of Science, PubPsych, and PsycInfo. Out of 670 initial results, 28 studies met the inclusion criteria and were selected for a detailed analysis. A qualitative thematic analysis of the investigated constructs identified six major themes in the literature, of which the theme Knowledge and Awareness was significantly underexplored, especially in studies involving non-academic populations. The results suggest that many laypersons lacked understanding of the capabilities and limitations of these tools, holding misconceptions such as the attribution of human-like language comprehension, the assumption that output quality depends solely on training data, or the—at the time incorrect—assumption that the systems can retrieve information in real time. Overall, we find a consistent pattern of overestimation of chatbot capabilities due to limited understanding. We lack studies testing whether myths about other AI systems also exist about AI-chatbots, as well as exploration of knowledge about AI-chatbots outside of academic samples and contexts. Building on a series of identified research gaps, we propose a research agenda that highlights the empirical work needed to better understand how the general public can be adequately informed about the use of AI chatbots.