Review systems in digital platforms help reduce the asymmetry of information between parts. These asymmetries may regard agents’ unobserved effort (Ebay), agents’ fixed characteristics (Amazon), or both (Airbnb) and they may lead to moral hazard and adverse selection issues in equilibrium. In this paper, I investigate how a review system is able to discipline these issues with an empirical estimate of the Airbnb hosts’ effort in each reviewed transaction. I do so by taking advantage of the multiple targets of the Airbnb reviews: with a sentiment analysis of the guests’ comments and ratings I measure the reviews’ content regarding the host behaviour and the dwelling characteristics. Next, I regress the host behaviour on dwelling characteristics contents with a panel fixed effect approach: with this procedure the effort the hosts exert in each transaction can be separated from the guests’ reviewing attitude. The effort estimates show that Airbnb hosts exert higher effort in the first comments and then shirk with a monotone decrease of the effort level until the exit. Changes in hosts’ effort lead to lagged changes of their turnover in the same direction, while prices remain fixed. I reinforce my identification strategy with a dynamic theoretical framework in line with all the previous empirical findings.