This thesis investigates the impact of a change introduced at a high-stake language exam in Hungary, following a paradigm shift concerning a test task eliciting an individual long turn from candidates. Originally, this task required test takers to produce a long turn and then to participate in interaction within the same task. Now, the task aims to assess production only, with awarding lower test scores concerning two components, namely Communicative value and Discourse management, of the rating scale if interaction is to take place. The findings contradict well-established results of the fluency-accuracy continuum research. What is more, significant changes in test scores can only be observed at the components in which no changes were made.