An essential component of seismic hazard analysis is the prediction of ground shaking (and its uncertainty), using ground motion models (GMMs). This paper proposes a new method to evaluate (i.e., rank) the suitability of GMMs for modelling ground motions in a given region. The method leverages a statistical tool from sensitivity analysis to quantitatively compare predictions of a GMM with underlying observations. We demonstrate the performance of the proposed method relative to several other popular GMM ranking procedures and highlight its advantages, which include its intuitive scoring system and its ability to account for the hierarchical structure of GMMs. We use the proposed method to evaluate the applicability of several GMMs for modelling ground motions from induced earthquakes due to UK shale gas exploration. The data consist of 195 recordings at hypocentral distances (R) less than 10 km for 29 events with local magnitude (ML) greater than 0 that relate to 2018/2019 hydraulic fracture operations at the Preston New Road shale gas site in Lancashire and 192 R < 10 km recordings for 48 ML > 0 events induced - within the same geologic formation - by coal mining near New Ollerton, North Nottinghamshire. We examine: (1) the Akkar et al. (2014a) models for European seismicity; (2) the Douglas et al. (2013) model for geothermal-induced seismicity; and (3) the Atkinson (2015) model for central and eastern North America induced seismicity. We find the Douglas et al. (2013) model to be the most suitable for almost all of the considered ground motion intensity measures. We modify this model by re-computing its coefficients in line with the observed data, to further improve its accuracy for future analyses of the seismic hazard of interest. This study both advances the state-of-the-art in GMM evaluation and enhances understanding of the seismic hazard related to UK shale gas exploration.