We analyze a data-set including more than 4.5 million tweets related to four highly emotional riot events. In particular, we examine statistically significant structural patterns that emerge as humans directly engage in an exchange of emotional messages with other humans on Twitter. Furthermore, we compare typical human-to-human communication patterns with those that emerge as bots engage in an emotional message-exchange with human users. To this end, we apply the novel concept of emotion-exchange motifs. We found that a) human-to-human conversations results in a variety of motifs that contain reciprocal edges and self-loops, b) bots predominantly contribute to the emergence of message broadcasting, single-way message sending behavior, c) in contrast to previous findings we found that in certain events bots frequently engage in direct message exchanges with humans, d) during riot events bots tend to direct fear-conveying messages to human users.