Geography and availability of natural habitat determine whether cropland intensification or expansion is more detrimental to biodiversity.

To mitigate biodiversity loss from agriculture, intensification is often promoted as an alternative to farmland expansion. However, its local impacts remain debated. We assess globally the responses of three biodiversity metrics-species richness, total abundance and relative community abundance-weighted average range size (RCAR), a proxy for biotic homogenization-to land conversion and yield increases. Our models predict a median species loss of 11% in primary vegetation in modified landscapes, and of 25% and 40% in cropland within natural and modified landscapes, respectively. Land conversion also reduces abundance and increases biotic homogenization, with impacts varying by geographic region and history of human modification. However, increasing yields changes biodiversity as well, including in adjacent primary vegetation, with effects dependent on crop, region, biodiversity metric and natural habitat cover. Ultimately, neither expansion nor intensification consistently benefits biodiversity. Intensification has better species richness outcomes in 29%, 83%, 64% and 57% of maize, soybean, wheat and rice landscapes, respectively, whereas expansion performs better in the remaining areas. In terms of abundance and RCAR, both expansion and intensification can outperform the other depending on landscape. Therefore, minimizing local biodiversity loss requires a context-dependent balance between expansion and intensification, while avoiding expansion in unmodified landscapes.