Resumes are often the first medium through which job applicants are evaluated, and resume screening could be the starting point for gender equality in various aspects. Little is known, however, about how women and men write differently in their resumes and how language differences are associated with the gender gap in labor. Here we analyze 6.7 million resumes of Chinese job applicants and find substantial gender resume differences, where women and men show distinct patterns in both simple language features and high-level semantic structures in the word embedding space of resumes. In particular, women tend to write shorter resumes but longer sentences and use a more diverse set of unique words, indicating their better language skills. Neural network models trained on resumes can predict gender with 80% accuracy, and the accuracy decreases with education levels and text standardization requirements. Moreover, while better language skills are associated with higher salary expectations, this positive relationship is magnified for men but weakened for women in women-dominated occupations. The results suggest that gender is deeply encoded in resumes, and language skills' benefits depend on occupational characteristics and gender. Our findings contribute to understanding gender gaps through the lens of written language in a resume.