Biodiversity indicators provide a vital window on the state of the planet, guiding policy development and management. The most widely adopted marine indicator is mean trophic level (MTL) from catches, intended to detect shifts from high-trophic-level predators to low-trophic-level invertebrates and plankton-feeders. This indicator underpins reported trends in human impacts, declining when predators collapse ("fishing down marine food webs") and when low-trophic-level fisheries expand ("fishing through marine food webs"). The assumption is that catch MTL measures changes in ecosystem MTL and biodiversity. Here we combine model predictions with global assessments of MTL from catches, trawl surveys and fisheries stock assessments and find that catch MTL does not reliably predict changes in marine ecosystems. Instead, catch MTL trends often diverge from ecosystem MTL trends obtained from surveys and assessments. In contrast to previous findings of rapid declines in catch MTL, we observe recent increases in catch, survey and assessment MTL. However, catches from most trophic levels are rising, which can intensify fishery collapses even when MTL trends are stable or increasing. To detect fishing impacts on marine biodiversity, we recommend greater efforts to measure true abundance trends for marine species, especially those most vulnerable to fishing.