Research relating to VUIs (voice user interface) such as voice assistants like Siri, Google Home, and Alexa has mostly been focused on general usage, error handling, and privacy concerns. This study focuses on mental models in relation to VUI - how users' presumptions about scope and how VUIs work match the reality. Specifically, this study aims to find out how mental models of VUIs form and evolve as well as whether or not mental models of VUIs affect usage. An online survey directed towards Google Home users showed its respondents, mainly identified as early adopters and primary users, showcasing a good mental model of Google Home together with frequent and varied usage as well as self-expressed knowledge. Due to suspected distortion of results caused by recruiting from an interest group on Facebook, study results are not considered to be representative to the population (VUI users in Sweden). This study showed that users base their mental models of VUIs around conventions found in human-to-human conversation and that these mental models evolve through feedback and error handling and can be improved by users exploring and testing more use cases for the VUI.