Reconstructing past seawater temperatures is essential for understanding ocean circulation and climate linkages, prompting extensive efforts to develop geochemical temperature proxies in marine carbonates that extend records beyond instrumental observations. Trace element ratios in biogenic carbonates offer valuable insights into past ocean temperatures, yet their reliability hinges on a detailed understanding of the chemical heterogeneity within marine calcifiers and the underlying biomineralization mechanisms. This study calibrates Li/Mg, Li/Ca, and Mg/Ca ratios in the corallite (cup skeleton) and branch (i.e., coenosteum) of modern colonial cold-water scleractinian corals (Madrepora, Enallopsammia, Dendrophyllia, Solenosmilia), and uses a numerical biomineralization model to investigate the influence of physiological processes and growth rates on these proxies. We find a systematic offset of Li/Ca, Mg/Ca as well as Li/Mg between corallite and branch, with higher values observed in the corallite. Our results show that while a biocalcification model that assumes constant metal (Me) distribution coefficients (DMe = Me/Cacarbonate / Me/Cafluid) effectively captures the observed Me/Ca correlations, it fails to explain the consistent Li/Mg offset between corallite and branch in these colonial coral specimens. Instead, growth rate difference between corallite and branch might contribute to the consistent Li/Mg offset between the two skeleton structures, with lower growth rates corresponding to the branch. This insight is critical for palaeotemperature reconstructions using colonial corals, as applying a single Li/Mg-temperature calibration to both skeletal components can introduce systematic error. We highlight the need to differentiate between skeletal structures, especially in fossil materials where corallites are often degraded. Our study underscores the need for continued research to reduce the uncertainties associated with Li/Mg paleothermometry.