Does Twitter trigger negativity in politicians digital communication? On social media direct feedback mechanisms like retweets or likes signal to politicians which message and tone are popular. Current research suggests that negative language increases the number of retweets a single tweet receives, indicating preferences for negativity in the audience on Twitter. However, it remains unclear whether politicians adapt to the platform logic of Twitter or simply follow the rules determined by the broader political context, namely the state of their electoral race. We use sentiment analysis to measure the tone used by 342 candidates in 97,909 tweets in their Twitter campaign in the 2018 midterm elections for the U.S. House of Representatives. We map the ideological structure of each politician’s filter bubble using IRT scaling of the her follower network. We can show that the feedback that candidates receive creates an incentive to use negativity. The size and direction of the tonal incentive depends on the ideological composition of the candidate’s filter bubble. Unexpectedly, the platform-specific incentive does not affect the tone used by candidates in their Twitter campaigns. Instead we find that the tone is affected by characteristics of the electoral race. We show that our findings are not dependent on our sentiment measurement by validating our results using hand coding and machine learning.