The loss and degradation of mangrove forests have triggered global restoration efforts to support biodiversity and ecosystem services, including fish stock enhancement. As mangrove restoration accelerates, it is important to evaluate outcomes for species that play functional roles in ecosystems and support services, yet this remains a clear knowledge gap. There is remarkably little information, for example, about how fish use of mangroves varies as restored vegetation matures, hampering efforts to include fisheries benefits in natural capital assessments of restoration. We used unbaited underwater cameras within two distinct zones of mangrove forests-fringe and interior-at five pairs of restored-natural mangrove sites of increasing age from restoration in southeast Queensland, Australia. We used deep learning to automatically extract data for the four most common species: yellowfin bream (Acanthopagrus australis), sea mullet (Mugil cephalus), common toadfish (Tetractenos hamiltoni), and common silverbiddy (Gerres subfasciatus). The abundance of these species varied among sites and zones, but was equal or greater in restored sites compared to paired natural sites. Despite younger restored sites having dramatically lower structural vegetation complexity, abundances did not increase with restoration site maturity. Furthermore, while yellowfin bream and sea mullet were more abundant in the fringe zone, we observed similarities in how fish used fringe and interior zones across all sites. Our paired, space-for-time design provides a powerful test of restoration outcomes for fish, highlighting that even newly restored sites with immature vegetation are readily utilized by key fish species.