Abstract Species distribution models (SDMs) are an important tool for conservation and resource management. However, managers are often interested in derived quantities such as range or area occupied, and how these are calculated can have a large impact. Ecosystem‐based management typically requires spatial information about species distributions, which is increasingly generated from SDMs that are then processed to identify occupied habitat. Many types of SDMs exist, but there is little research regarding how this model‐choice affects outcomes when defining occupied habitat, in part because these models generate different types of output. We fit a suite of five SDMs to data for 208 species/life stage combinations in three marine ecosystems while ensuring that they all estimate a ‘common currency’ of numerical abundance. We then calculate out‐of‐sample predictive performance to weight these constituents in an ensemble SDM. Results show that this approach can reduce bias arising from a priori specification of individual SDMs resulting in a better fit to survey data (constituent SDMs had a median of 7% higher RMSE). The SDMs had a range of responses relative to the ensemble, with MaxEnt typically predicting a median 1.3% higher area occupied, and negative‐binomial GAMs predicting 21.4% lower area occupied. Two potential methods of identifying the area of occupied habitat from SDM outputs are compared—probability‐based and cumulative density‐based methods. We find that cumulative densities result in smaller estimates of area occupied, and we recommend careful consideration of how model‐choice affects occupied‐habitat estimates in spatial management. Policy implications: Finally, we discuss how the patterns identified during the 5‐year Review of Essential Fish Habitat for Alaska should be carefully considered by managers using SDMs to identify habitat that may be impacted by anthropogenic activities.