An approach is provided that improves a question answering (QA) computer system by ingesting a set of documents into a corpus of a first QA system. A multiple choice test is generated from the set of documents. The multiple choice test includes any number of questions, with each of the questions having a correct answer and one or more candidate, or wrong answers. The questions and answers are retrieved from the set of documents. The approach identifies concepts that correspond to each of the questions. The candidate answers corresponding to each of the questions are generated based on the identified concepts that pertaining to the respective questions. In this manner, the candidate answers are designed to be misleading in light of the respective questions. A subject, such as a second QA system, is tested using the generated multiple choice test.