Estimates of deep-time biodiversity typically rely on statistical methods to mitigate the impacts of sampling biases in the fossil record. However, these methods are limited by the spatial and temporal scale of the underlying data. Here we use a spatially explicit mechanistic model, based on neutral theory, to test hypotheses of early tetrapod diversity change during the late Carboniferous and early Permian, critical intervals for the diversification of vertebrate life on land. Our simulations suggest that apparent increases in early tetrapod diversity were not driven by local endemism following the ‘Carboniferous rainforest collapse’. Instead, changes in face-value diversity can be explained by variation in sampling intensity through time. Our results further demonstrate the importance of accounting for sampling biases in analyses of the fossil record and highlight the vast potential of mechanistic models, including neutral models, for testing hypotheses in palaeobiology.