To Discriminate, or Not to Discriminate, in Online Job Ads

How do employers use and adjust recruitment signals and wage offers when they target applicants of a particular gender? Using online vacancy data from Vietnam and natural language processing methods, we construct an index measuring the direction and intensity of implicit gendered language in job requirements. We document widespread implicit gendered framing and substantial variation across repeated postings for the same firm and job, even where explicit gender requests remain unchanged. When explicit requests do change, we find that the two signals move together: adding a gender request strengthens wording oriented toward that gender, while removing it weakens such wording. Both signals are also associated with posted wages. Within positions, adding a male request is accompanied by 3% higher wages, and adding a female request by 4% lower wages. More male-coded wording is likewise associated with higher wages. A stylised directed-search model can rationalise these patterns when applicants' wage responsiveness differs by gender and inconsistent recruitment signals discourage applicants.