The splitting of water photoelectrochemically into hydrogen and oxygen represents a promising technology for converting solar energy to fuel(1,2). The main challenge is to ensure that photogenerated holes efficiently oxidize water, which generally requires modification of the photoanode with an oxygen evolution catalyst (OEC) to increase the photocurrent and reduce the onset potential(3). However, because excess OEC material can hinder light absorption and decrease photoanode performance(4), its deposition needs to be carefully controlled-yet it is unclear which semiconductor surface sites give optimal improvement if targeted for OEC deposition, and whether sites catalysing water oxidation also contribute to competing charge-carrier recombination with photogenerated electrons(5). Surface heterogeneity(6) exacerbates these uncertainties, especially for nanostructured photoanodes benefiting from small charge-carrier transport distances(1,7,8). Here we use super-resolution imaging(9-13), operated in a charge-carrier-selective manner and with a spatiotemporal resolution of approximately 30 nanometres and 15 milliseconds, to map both the electron-and hole-driven photoelectrocatalytic activities on single titanium oxide nanorods. We then map, with sub-particle resolution (about 390 nanometres), the photocurrent associated with water oxidation, and find that the most active sites for water oxidation are also the most important sites for charge-carrier recombination. Site-selective deposition of an OEC, guided by the activity maps, improves the overall performance of a given nanorod-even though more improvement in photocurrent efficiency correlates with less reduction in onset potential (and vice versa) at the sub-particle level. Moreover, the optimal catalyst deposition sites for photocurrent enhancement are the lower-activity sites, and for onset potential reduction the optimal sites are the sites with more positive onset potential, contrary to what is obtainable under typical deposition conditions. These findings allow us to suggest an activity-based strategy for rationally engineering catalyst-improved photoelectrodes, which should be widely applicable because our measurements can be performed for many different semiconductor and catalyst materials.