In a type of interactions over the Internet, a user (the subject) is dependent on another user (the target), but not vice versa. The subject should therefore form an opinion about the target, before possibly initiating an interaction. The scenario wherein a subject only relies on information obtained from past interactions with the target, is well-studied and understood. In this paper, we formally analyze the implication of allowing recommendations (statements of a third party) as source of information. We identify the family of valid models that admit recommendations. This allows us to verify particular existing models that admit recommendations.