Recently, the Greenland Ice Sheet (GrIS) has become the main source of barystatic sea-level rise. The increase in the GrIS melt is linked to anticyclonic circulation anomalies, a reduction in cloud cover and enhanced warm-air advection. The Climate Model Intercomparison Project fifth phase (CMIP5) General Circulation Models (GCMs) do not capture recent circulation dynamics; therefore, regional climate models (RCMs) driven by GCMs still show significant uncertainties in future GrIS sea-level contribution, even within one emission scenario. Here, we use the RCM Modèle Atmosphèrique Règional to show that the modelled cloud water phase is the main source of disagreement among future GrIS melt projections. We show that, in the current climate, anticyclonic circulation results in more melting than under a neutral-circulation regime. However, we find that the GrIS longwave cloud radiative effect is extremely sensitive to the modelled cloud liquid-water path, which explains melt anomalies of +378 Gt yr–1 (+1.04 mm yr–1 global sea level equivalent) in a +2 °C-warmer climate with a neutral-circulation regime (equivalent to 21% more melt than under anticyclonic circulation). The discrepancies between modelled cloud properties within a high-emission scenario introduce larger uncertainties in projected melt volumes than the difference in melt between low- and high-emission scenarios.