Studying the prevalence of putatively rare behaviours, such as tool use, is challenging because absence of evidence can arise either from a species' inability to produce the behaviour or from insufficient research effort. Here, we tackle this challenge by combining crowdsourcing and phylogenetic modelling to approximate actual rates of a rarely observed behaviour based on limited data, targeting tool use in parrots. Crowdsourcing on a social media platform revealed novel instances of tool use in 17 parrot species, more than doubling the confirmed number of tool-using parrot species from 11 (3%) to 28 (7%). Phylogenetic modelling ranked additional species that are most likely to be unobserved tool users, suggesting that between 11% and 17% of extant parrot species may be tool users. These discoveries have implications for inferences about the evolutionary drivers and origins of tool use in parrots, revealing associations with relative brain size and feeding generalism and indicating several genera where tool use was likely an ancestral trait. Overall, our findings challenge the assumption that current sampling effort captures the full distribution of putatively rare animal behaviours. Combining our sampling and analysis methods offers a fruitful approach for investigating the distribution, drivers, and origins of other rare behaviours.