Anisotropies of the cosmic microwave background (CMB) have proven to be a very powerful tool to constrain dark matter annihilation at the epoch of recombination. However, CMB constraints are currently derived using a number of reasonable but yet untested assumptions that could potentially lead to a misestimation of the true bounds (or any reconstructed signal). In this paper we examine the potential impact of these systematic effects. In particular, we separately study the propagation of the secondary particles produced by annihilation in two energy regimes: first following the shower from the initial particle energy to the keV scale, and then tracking the resulting secondary particles from this scale to the absorption of their energy as heat, ionization, or excitation of the medium. We improve both the high- and low-energy parts of the calculation, in particular finding that our more accurate treatment of losses to sub-10.2 eV photons produced by scattering of high-energy electrons weakens the constraints on particular dark matter annihilation models by up to a factor of 2. On the other hand, we find that the uncertainties we examine for the low-energy propagation do not significantly affect the results for current and upcoming CMB data. We include the evaluation of the precise amount of excitation energy, in the form of Lyman-alpha photons, produced by the propagation of the shower, and examine the effects of varying the helium fraction and helium ionization fraction. In the recent literature, simple approximations for the fraction of energy absorbed in different channels have often been used to derive CMB constraints: we assess the impact of using accurate vs approximate energy fractions. Finally we check that the choice of recombination code (between RECFAST V1.5 and COSMOREC), to calculate the evolution of the free electron fraction in the presence of dark matter annihilation, introduces negligible differences.