Seismic hazard analyses (SHA) are routinely carried out around the world to understand the hazard, and consequently the risk, posed by earthquake activity. Whether single scenario, deterministic analyses, or state-of-the art probabilistic approaches, considering all possible events, a founding pillar of SHA is the estimation of the ground-shaking field from potential future earthquakes. Early models accounted for simple observations, such that ground shaking from larger earthquakes is stronger and that ground motion tends to attenuate rapidly away from the earthquake source. The first ground motion prediction equations (GMPEs) were, therefore, developed with as few as two principal predictor variables: magnitude and distance. Despite the significant growth of computer power over the last few decades, and with it the possibility to compute kinematic or dynamic rupture models coupled with simulations of 3D wave propagation, the simple parametric GMPE has remained the tool of choice for hazard analysts. There are numerous reasons for this. First and foremost GMPEs are robust and reliable within the model space considered during their derivation, and many can be extrapolated to a degree beyond this space with some confidence. With ever expanding datasets and improved metadata the models are becoming more and more useful: a range of predictor variables are now used, describing the source, path and site effects in detail. GMPEs are also relatively easy to implement and computationally inexpensive. Despite this, probabilistic hazard calculations using GMPEs and accounting for uncertainties can still take several days to run. Full simulation-based approaches, therefore, clearly lie outside the computation budget afforded to most projects. As well as the ever expanding list of predictor variables, other recent developments have also significantly improved the predictive power of GMPEs. This has allowed them to maintain their advantage over more `physical' simulation techniques. Possibly the biggest aspect of this is not related to the median ground-shaking field, but rather its variability (and correlation in space and with oscillator period). This is a major advantage of empirical as opposed to simulation approaches, which typically struggle to replicate the covariance of input variables and, consequently, the variance of the ground motion. In this article we summarize some of the recent advances in ground motion prediction equations, including their application in SHA. We begin with a summary of the current state-of-the-art, then introduce the main additional predictor variables now used. Region- and event-type (tectonic or induced) specific predictions and adjustments are then discussed. Additional topics include advances in estimating ground-motion variability (epistemic and aleatory) and expanding GMPEs to predict other intensity measures or waveform features. The article concludes with a discussion on the path forward in earthquake ground motion prediction