Recommender systems have their origin in e-commerce. In this domain the users are meant to like the recommended information. This preference-consistency is not adequate or even desirable for all domains where recommender systems are implemented. One key issue for opinion formation and informed decision making is to be aware of more than one's own perspective. However, information search is often biased, because confirming information is favored over opposing information. Therefore it would be useful to recommend information that is inconsistent to users' prior perspective to help overcome this bias. The present paper deals with an online experiment aimed at investigating the effects of preference-consistent compared to preference-inconsistent recommendations on information selection and evaluation. Results showed a significant reduction of confirmation bias in the condition with preference-inconsistent recommendations. However, participants prefer preference-consistent recommendations in terms of global, cognitive and affective evaluations. We discuss the impact of these findings for application.