This paper studies the production of offensive content by politicians on Twitter/X. Applying AI-based methods to the universe of tweets posted by members of the U.S. Congress (2017-2022), I measure the returns to offensive communication both in terms of online engagement and electoral support. I document that posting toxic tweets generates a sizable engagement premium and that these gains decrease with politicians' baseline toxicity. To examine how voters respond to offensive speech, I link the timing of toxic tweets to a large-scale electoral survey. Using an event study design around days marked by sharp spikes in representatives’ toxic tweeting activity, I find that voting intentions for the politician increase in the following week. The increase is concentrated among ideologically aligned voters, while opponents display lower electoral support, widening the partisan voting gap. These findings highlight the presence of both online and electoral incentives to the production of offensive communication, which comes at the expense of growing polarization.