Processes involving heavy quarks are a crucial component of the LHC physics program, both by themselves and as backgrounds for Higgs physics and new physics searches. In this work, we critically reconsider the validity of the widely-adopted approximation in which heavy quarks are generated at the matrix-element level, with special emphasis on the impact of the collinear logarithms associated with final-state heavy quark and gluon splittings. Our study, based on a perturbative fragmentation-function approach, explicitly shows that neglecting the resummation of collinear logarithms may yield inaccurate predictions, in particular when observables exclusive in the heavy quark degrees of freedom are considered. Our findings motivate the use of schemes which encompass the resummation of final-state collinear logarithms.