We make two contributions to the literature exploring the role of sentiment in macroeconomic fluctuations: (I) Working with the theoretical MA representations of standard DSGE models, we show that several SVAR-based approaches to the identification of sentiment shocks are unreliable, as (e.g.) they identify such disturbances even when the model does not feature them. The approach proposed by Beaudry et al. (2011), for example, identifies sentiment shocks within Smets and Wouters' (2007) model. The problem is that the restrictions which are typically imposed are so weak and generic that they will always be satisfied with non-negligible probability by random rotations of the model's structural disturbances, irrespective of the fact that they do, or do not include a pure sentiment shock. (II) We derive robust restrictions for the identification of sentiment shocks based on the model of Angeletos et al. (2018), and working with the theoretical MA representation of the model we show that they allow to recover the shocks' IRFs and fractions of forecast error variance either exactly, or with great precision. When we impose these restrictions upon the data within a structural VAR framework, we consistently detect a minor-to-negligible role for sentiment shocks in business-cycle fluctuations.