Do female and male candidates campaign differently, and why? Using individual political platforms from French legislative elections between 1981 and 2024, I combine a transformer machine learning model with a regression discontinuity design that exploits the two-round structure of French elections to identify the causal impact of candidate gender on campaign content. Female and male candidates who barely qualify discuss the same issues in the first-round, but women write 6 percentage points less of their platform in a communal register. Between the two rounds, the gender gap in male-stereotyped topics rises by 4.7 percentage points, driven by male candidates, who withdraw from these topics and reallocate space to framing the national contest. This effect appears only when male politicians compete against another man, as they try to differentiate themselves from their opponents. Women increase their use of communal language, closing the gender gap, but do not adapt their behavior to their opponent's gender.