This paper deals with tag questions to which an answer is demanded by a speaker who is certain about the truth of the proposition but who wants to hear the answer uttered by the addressee. Similar tag questions have previously been described based on data from courtrooms (e.g. Biscetti 2006), where tag questions are typically used by powerful speakers. However, data from the British National Corpus shows that confirmation-demanding tag questions may also be used outside institutional settings and in situations with various power relationships. Most of these examples are from fiction dialogue, where conflicts and confrontations are often depicted. In courtrooms, there is always an audience; however, in fiction dialogue, most confirmation-demanding questions in the data are found in private conversations. Confirmation-demanding tag questions seldom seem to be captured in conversational data, apart from in cases where the speaker wants the answer to be heard by a third party; it is therefore suggested that private confrontations might be underrepresented in conversational data. This paper also discusses functional categorizations of tag questions in general and argues that the unit of analysis should be the whole tag question, i.e. the anchor and the tag taken together, and not just the tag.