This paper presents a method to estimate political ideology based on the the words and phrases candidates use on their websites and Twitter. The method enables estimation for both incumbents and challengers and, unlike methods based on campaign contributions or follower counts, the method is built on behavior under the control of candidates that directly relates to how they are perceived by voters. In contrast to existing approaches, the approach can accommodate important heterogeneity across parties in the ideological implications of many terms (i.e., liberal Democrats and conservative Republicans both use terms such as "black lives matter" or "President Trump" more than moderates). The process produces term parameter and ideal point estimates that have high face, convergent and construct validity. To demonstrate the utility of these estimates, this paper presents evidence that ideological moderation was electorally beneficial in the 2020 congressional general election.