We developed a quiz game that uses questions automatically generated from the Czech and English DBpedia. The game uses class membership and categorization to split DBpedia resources into domains of questions, such as persons or films. Difficulty of questions is estimated by using the indegree graph metric based on links between Wikipedia pages. Surface forms of the DBpedia resources including inbound link labels and select properties are used to remove potential spoilers from questions. We evaluate the relation of indegree to answer success rate and the contribution of semantic features for spoiler removal using online evaluation during game sessions with actual users. Our findings indicate that answer success rate correlates with indegree, but it has a stronger correlation with page views. The evaluation also suggests that additional surface forms improve spoiler detection.