Can Google trends be used to estimate the geographic distribution of alien plants in the United States?

Google Trends has long been used to track the epidemiology of human diseases; however, its application to address biological invasions has been quite limited to date. We develop a workflow for best practice in the use of Google Trends to study biological invasions that accounts for the underlying pitfalls inherent in data from Google searches. We illustrate this workflow by examining the extent Google searches adequately depict the state-wide occurrence of 100 alien plant species in the United States. Google Trends outputs are based on samples of search queries in the Google search engine during a specific period, and thus, despite many studies using results from a single Google Trends search, we show that it is essential to undertake multiple replicate searches for robust interpretation. In general, results from Google Trends provided only a moderate goodness of fit to the known state-wide occurrence of alien plant species, and then only when the scientific name was used as a keyword. Other keywords, such as the common name or a Topic query, performed poorly. The goodness of fit between the observed species occurrence and that predicted using Google searches was higher for ornamental species or those officially classed by the USDA as noxious or invasive in at least one state. Those states with a greater alien plant richness and higher average education level of their citizens performed best. Given this data heterogeneity, we highlight an objective workflow to assess the value of Google Trends in order to ensure that greater scrutiny is applied when using this tool in invasion science. • A robust workflow for using Google Trends to study biological invasions is developed. • Scientific names as keywords best reflect alien plant species occurrence in US states. • Accuracy was highest for ornamental plants or well documented invasive species. • States with a larger proportion of higher educated residents performed best. • Results were not consistent enough for prediction or surveillance at a state-level.