One of the most important biological processes at the molecular level is the formation of protein-ligand complexes. Therefore, determining their structure and underlying key interactions is of paramount relevance and has direct applications in drug development. Because of its low cost relative to its experimental sibling, molecular dynamics (MD) simulations in the presence of different solvent probes mimicking specific types of interactions have been increasingly used to analyze protein binding sites and reveal protein-ligand interaction hot spots. However, a systematic comparison of different probes and their real predictive power from a quantitative and thermodynamic point of view is still missing. In the present work, we have performed MD simulations of 18 different proteins in pure water as well as water mixtures of ethanol, acetamide, acetonitrile and methylammonium acetate, leading to a total of 5.4 μs simulation time. For each system, we determined the corresponding solvent sites, defined as space regions adjacent to the protein surface where the probability of finding a probe atom is higher than that in the bulk solvent. Finally, we compared the identified solvent sites with 121 different protein-ligand complexes and used them to perform molecular docking and ligand binding free energy estimates. Our results show that combining solely water and ethanol sites allows sampling over 70% of all possible protein-ligand interactions, especially those that coincide with ligand-based pharmacophoric points. Most important, we also show how the solvent sites can be used to significantly improve ligand docking in terms of both accuracy and precision, and that accurate predictions of ligand binding free energies, along with relative ranking of ligand affinity, can be performed.