From search results to generative answers: A mixed-methods study of credibility assessment and trust formation in AI Overviews

Abstract

Generative search shifts information evaluation from comparing ranked documents to assessing synthesised answers. This study examines how users’ perceptions of answer quality are associated with trust and credibility in Google AI Overviews and explores whether visible interface cues may foster trust without systematic source verification. Study 1 analysed survey data from 510 Spanish users using a consistent partial least squares structural equation modelling approach. Study 2 combined eye tracking and interviews with 20 Spanish users. The model explained 75.7% of the variance in trust and 91.0% of the variance in credibility. Clarity and context ( β = .419, p < .001) and access to additional information and supporting sources ( β = .335, p = .002) showed the strongest observed associations with trust, whereas response immediacy was not significant ( β = .008, p = .874). Trust ( β = .834) and perceived usefulness ( β = .229) were positively associated with credibility (both p < .001), which, in turn, was associated with recommendation following ( β = .877) and intention to use ( β = .553; both p < .001). Eye tracking recorded no analysable fixations on the AI-error warning, while the interviews indicated that participants in this subsample did not routinely verify the displayed sources and often interpreted visible references as indicators of reliability. Taken together, Study 1 identifies associations between perceived answer-quality cues, trust, and credibility, whereas Study 2 suggests that visual prominence and visible references appeared to operate as authority cues for some participants and may encourage reliance without active source verification.