Anonymizing a job application doesn't work because your vocabulary gives your gender away.
April 17, 2026
Original Paper
Identifying and Mitigating Gender Cues in Academic Recommendation Letters: An Interpretability Case Study
arXiv · 2604.12337
The Takeaway
We’ve tried to fix hiring bias by removing names and pronouns from resumes, but AI can still guess the applicant's gender with 68% accuracy. It turns out that the words people use in recommendation letters—calling someone 'emotional' vs 'ambitious'—are incredibly reliable gender proxies. Even when we try to be professional, we use a gendered vocabulary of praise that AI picks up on instantly. This means that 'blind' hiring isn't actually blind; it’s just filtering for the same old biases through a different lens. For job seekers, it’s a reminder that the very way people describe your strengths might be holding you back.
From the abstract
Letters of recommendation (LoRs) can carry patterns of implicitly gendered language that can inadvertently influence downstream decisions, e.g. in hiring and admissions. In this work, we investigate the extent to which Transformer-based encoder models as well as Large Language Models (LLMs) can infer the gender of applicants in academic LoRs submitted to an U.S. medical-residency program after explicit identifiers like names and pronouns are de-gendered. While using three models (DistilBERT, RoB