Predicting the binding poses of ligands targeting RNAs is challenging. Here, we propose that using first-principles quantum mechanics/molecular mechanics (QM/MM) simulations, which incorporate automatically polarization effects, can help refine the structural determinants of ligand/RNA complexes in aqueous solution. In fact, recent advances in massively parallel computer architectures (such as exascale machines), combined with the power of machine learning, are greatly expanding the domain of applicability of these types of notoriously expensive simulations. We corroborate this proposal by carrying out a QM/MM-based study on a ligand targeting CAG repeat-RNA, involved in Huntington's disease. The calculations indeed show a clear improvement in the ligand binding properties, and they are consistent with the NMR measurements, also performed here. Thus, this type of approach may be useful for practical applications in the design of ligands targeting RNA in the near future.