Researchers have found that individuals in online social networks become more resolute in their beliefs when shown opinions opposite their own, a phenomenon referred to as the backfire effect. In this work we present a persuasion technique we call pacing and leading, which can mitigate the backfire effect. This dynamic technique has one gradually evolve their opinion over time to initially pace the persuasion target, and then lead them to the desired opinion. To test this technique, we conduct a field experiment in Twitter using artificial bot accounts where the goal is to persuade users who oppose immigration to support it. Neural network based sentiment analysis of the experimental subjects' tweets identifies a strong negative sentiment for tweets containing the word ``illegals''. Based on this, we use the frequency of tweets containing illegals as a measure of the treatment effect. We find that pacing and leading is more effective than simply presenting an opposing view, which we refer to as arguing, which actually results in a backfire effect. Our results suggest that dynamic persuasion techniques can be more effective than static approaches which present a constant opinion.