Plants modulate their rhizochemistry, which affects soil bacterial communities and, ultimately, plant performance. Although our understanding of rhizochemistry is growing, knowledge of its responses to abiotic constraints is limited, especially in realistic ecological contexts. Here, we combined predictive metabolomics with soil metagenomics to investigate how rhizochemistry responded to environmental constraints and how it in turn shaped soil bacterial communities across stress gradients in the Atacama Desert. We found that rhizochemical adjustments predicted the environment (i.e. elevation, R2 between 96% and 74%) of two plant species, identifying rhizochemical markers for plant resilience to harsh edaphic conditions. These metabolites (e.g. glutamic and succinic acid, catechins) were consistent across years and could predict the elevation of two independent plant species, suggesting biochemical convergence. Next, convergent patterns in the dynamics of bacterial communities were also observed across the elevation gradient. Finally, rhizosphere predictors were associated with variation in composition and abundance of bacterial species. Biochemical markers and convergences as well as potential roles of associated predictive bacterial families reflected the requirements for plant life under extreme conditions. This included biological processes such as nitrogen and water starvation (e.g. glutamic and organic acids, Bradyrhizobiaceae), metal pollution (e.g. Caulobacteraceae) and plant development and defence (e.g. flavonoids, lipids, Chitinophagaceae). Overall, findings highlighted convergent patterns belowground, which represent exciting insights in the context of evolutionary biology, and may indicate unique metabolic sets also relevant for crop engineering and soil quality diagnostics. Besides, the results emphasise the need to integrate ecology with omics approaches to explore plant-soil interactions and better predict their responses to climate change.