Graphitic carbon nitride (g-C3N4) has recently emerged as a promising visible-light-responsive polymeric photocatalyst; however, a molecular-level understanding of material properties and its application for water purification were underexplored. In this study, we rationally designed nonmetal doped, supramolecule-based g-C3N4 with improved surface area and charge separation. Density functional theory (DFT) simulations indicated that carbon-doped g-C3N4 showed a thermodynamically stable structure, promoted charge separation, and had suitable energy levels of conduction and valence bands for photocatalytic oxidation compared to phosphorus-doped g-C3N4. The optimized carbon-doped, supramolecule-based g-C3N4 showed a reaction rate enhancement of 2.3-10.5-fold for the degradation of phenol and persistent organic micropollutants compared to that of conventional, melamine-based g-C3N4 in a model buffer system under the irradiation of simulated visible sunlight. Carbon-doping but not phosphorus-doping improved reactivity for contaminant degradation in agreement with DFT simulation results. Selective contaminant degradation was observed on g-C3N4, likely due to differences in reactive oxygen species production and/or contaminant-photocatalyst interfacial interactions on different g-C3N4 samples. Moreover, g-C3N4 is a robust photocatalyst for contaminant degradation in raw natural water and (partially) treated water and wastewater. In summary, DFT simulations are a viable tool to predict photocatalyst properties and oxidation performance for contaminant removal, and they guide the rational design, fabrication, and implementation of visible-light-responsive g-C3N4 for efficient, robust, and sustainable water treatment.