Aim Predicting the spatial distribution of species assemblages remains an important challenge in biogeography. Recently, it has been proposed to extend correlative species distribution models (SDMs) by taking into account (a) covariance between species occurrences in so‐called joint species distribution models (JSDMs) and (b) ecological assembly rules within the SESAM (spatially explicit species assemblage modelling) framework. Yet, little guidance exists on how these approaches could be combined. We, thus, aim to compare the accuracy of assemblage predictions derived from stacked and from joint SDMs. Location Switzerland. Taxon Birds, tree species. Methods Based on two monitoring schemes (national forest inventory and Swiss breeding bird atlas), we built SDMs and JSDMs for tree species (at 100 m resolution) and forest birds (at 1 km resolution). We tested accuracy of species assemblage and richness predictions on holdout data using different stacking procedures and ecological assembly rules. Results Despite minor differences, results were consistent between birds and tree species. Cross‐validated species‐level model performance was generally higher in SDMs than JSDMs. Differences in species richness and assemblage predictions were larger between stacking procedures and ecological assembly rules than between stacked SDMs and JSDMs. On average, predictions were slightly better for stacked SDMs compared to JSDMs, probabilistic stacks outperformed binary stacks, and ecological assembly rules yielded best predictions. Main conclusions When predicting the composition of species assemblages, the choice of stacking procedure and ecological assembly rule seems more decisive than differences in underlying model type (SDM vs. JSDM). JSDMs do not seem to improve community predictions compared to SDMs or improve predictions for rare species. Still, JSDMs may provide additional insights into community assembly and may help deriving hypotheses about prevailing biotic interactions in the system. We provide simple rules of thumb for choosing appropriate modelling pathways. Future studies should test these preliminary guidelines for other taxa and biogeographical realms as well as for other JSDM algorithms.

based on data of limited sample size or limited regional coverage. We thus recommend home range scale analyses and sampling designs that cover diverse regional landscapes and ensure robust estimates of habitat suitability to conserve wild animal populations.