Recommender systems are present in many web applications to guide purchase choices. They increase sales and benefit sellers, but whether they benefit customers by providing relevant products remains less explored. While in many cases the recommended products are relevant to users, in other cases customers may be tempted to purchase the products only because they are recommended. Here we introduce a model to examine the benefit of recommender systems for users, and find that recommendations from the system can be equivalent to random draws if one always follows the recommendations and seldom purchases according to his or her own preference. Nevertheless, with sufficient information about user preferences, recommendations become accurate and an abrupt transition to this accurate regime is observed for some of the studied algorithms. On the other hand, we find that high estimated accuracy indicated by common accuracy metrics is not necessarily equivalent to high real accuracy in matching users with products. This disagreement between estimated and real accuracy serves as an alarm for operators and researchers who evaluate recommender systems merely with accuracy metrics. We tested our model with a real dataset and observed similar behaviors. Finally, a recommendation approach with improved accuracy is suggested. These results imply that recommender systems can benefit users, but the more frequently a user purchases the recommended products, the less relevant the recommended products are in matching user taste.