Pedogenic carbonate is an invaluable archive for reconstructing continental paleoclimate and paleoecology. The δ13C of pedogenic carbonate (δ13Cc) has been widely used to document the rise and expansion of C4 plants over the Cenozoic. This application requires a fundamental presumption that in soil pores, soil-respired CO2 dominates over atmospheric CO2 during the formation of pedogenic carbonates. However, the decoupling between δ13Cc and δ13C of soil organic matter (δ13CSOM) have been observed, particularly in arid regions, suggesting that this presumption is not always valid. To evaluate the influence of atmospheric CO2 on soil δ13Cc, here we performed systematic δ13C analyses of paleosols across the Chinese Loess Plateau, with the sample ages spanning three intervals: the Holocene, the Late Pleistocene, and the mid-Pliocene warm period. Our paired δ13Cc and δ13CSOM data reveal broadly divergent trending patterns. Using a two-component CO2-mixing model, we show substantial incorporations of atmospheric CO2 (up to 60%) into soil pore space during carbonate precipitation. This result readily explains the enrichment of δ13Cc and its divergence from δ13CSOM. As a consequence, δ13C of pedogenic carbonates formed under semiarid and/or arid conditions are largely driven by regional aridity through its control on soil CO2 composition, and thus cannot be used to evaluate the relative abundance of C3 versus C4 plants. Nonetheless, these carbonates can be applied for atmospheric CO2 reconstructions, even for periods with low CO2 levels.