Social networking has a powerful influence on voters, but we do not know enough about the mechanisms of network influence. Recent research shows that one network member's influence is highly dependent on the others in the network, i.e. autoregressive. I test whether the influence of social network political knowledge is also autoregressive. I show that a strong predictor of vote choice similarity is the level of knowledge of the discussant, but greater knowledge of the other network members lessens dyadic agreement. Data from the American National Election Study collected in 2000 show that in the presidential election of 2000 having a knowledgeable discussant increases the chance of vote similarity with that discussant by 5 percentage points, but vote similarity decreases by 10 percentage points for each level of residual network knowledge. This research confirms the autoregressive influence of social network political knowledge.