Formative feedback currently receives attention as an effective means of increasing student learning. However, how to frame feedback to achieve the best effect is an ongoing debate. In this study we analyse a written data-set of 174 segments of teacher feedback and student response, coded using 10 emergent feedback and 14 response categories. As it is argued that feedback is a dialogue between students and teacher, links between feedback and response segments are viewed as a dialogical framework that we visualise and understand using network analysis. From this network we conclude that some ways of formulating feedback are more productive and likely to lead to types of responses that signify learning than others. We thus identify the reflection group of responses, consisting of the categories "reflective response," "explanation" and "students investigate own thinking." The feedback categories that link primarily to the reflection group are "open question," "wondering question" and "leading question", which we categorise as the "questioning group" of feedback. We discuss these patterns against a previously published framework, and by discussing specific examples we further our understanding of what makes feedback formative.