Graphitic carbon nitride has long been considered incapable of splitting water molecules into hydrogen and oxygen without adding small molecule organics despite the fact that the visible-light response and proper band structure fulfills the proper energy requirements to evolve oxygen. Herein, through in-situ observations of a collective C = O bonding, we identify the long-hidden bottleneck of photocatalytic overall water splitting on a single-phased g-C 3 N 4 catalyst via fluorination. As carbon sites are occupied with surface fluorine atoms, intermediate C=O bonding is vastly minimized on the surface and an order-of-magnitude improved H 2 evolution rate compared to the pristine g-C 3 N 4 catalyst and continuous O 2 evolution is achieved. Density functional theory calculations suggest an optimized oxygen evolution reaction pathway on neighboring N atoms by C-F interaction, which effectively avoids the excessively strong C-O interaction or weak N-O interaction on the pristine g-C 3 N 4 .