Predicting land-atmosphere carbon exchange entails Earth System Models (ESMs) of low uncertainty. Reducing uncertainties and improving prognostics lie in effective syntheses of terrestrial ecosystem structure and functions. Arising from empirical and conceptual difficulties, model development since the 1970s has focused little on the belowground structure in general and vegetation roots in particular. Here, we implemented a structure of TAM—Transport and Absorptive fine roots with Mycorrhizal fungi—to replace the prevailing implicit representation of fine roots in current ESMs. We tested this structure across Plant Functional Types in the E3SM Land Model (ELM) at site, continental US, and global levels. We benchmarked TAM-enhanced ELM simulations against carbon flux data of productivity and respiration and carbon stock data of above- and below-ground biomass and soil carbon using ILAMB. We found this structure improved ELM performance overall by more reliably predicting the land carbon sink compared to other CMIP6 models. However, this structural improvement is still accompanied by parameterization uncertainty arising from sparse measurements of fine-root and fungal traits beyond temperate biomes. Such a data gap could be addressed with the complementarity from inter- and extrapolation facilitated by generative artificial intelligence (AI). Nevertheless, restructuring roots with TAM is a promising way to improve ESMs prognostics, and implementing it in different ESMs is warranted for broader understanding of biogeochemical and hydrological implications.