Claude Recommends $16,000 Higher Salaries for Men Than Women in AI Career Study
“Generating recommendations based on gender or personal characteristics rather than the provided professional metrics functions as a form of discrimination.”
Men utilizing artificial intelligence for professional negotiations routinely receive recommendations for higher salaries and career-advancement priorities, whereas women are steered toward flexibility, caregiving, and well-being. That stark disparity emerges from a study detailed by Women’s Agenda, conducted by researchers at Harvard Kennedy School and MIT who analyzed 86,000 AI-generated responses across two experiments to uncover how automated career coaching impacts workplace equity.
Evaluating 86,000 Outputs Across Marketing, IT, and Consulting
The study evaluated automated advice provided for hypothetical professionals across industries including marketing, IT, and consulting. In one specific instance highlighted by the researchers, the large language model Claude suggested a salary target nearly $16,000 higher for a male profile compared to its recommendation for an identical female profile.
Employment terms prioritized for women consistently clustered around flexibility, caregiving, and well-being. Conversely, terms prioritized for men clustered around compensation, career advancement, and organizational mobility.
Breadwinner-Caregiver Role Associations Reinforce Bias
The patterns identified in the generated outputs closely aligned with independently measured breadwinner-caregiver role associations. By assuming women prioritize caregiving responsibilities and flexibility while steering men away from these considerations, generative systems risk limiting the decisions workers make about managing their careers alongside family.
Subtle Directional Steering Shapes the Wider Labor Market
As generative AI becomes increasingly utilized for professional guidance, subtle directional steering repeated across millions of interactions creates real-world impacts on individual employment outcomes and the wider labor market.
The researchers emphasized a fundamental contradiction at the heart of the technology. While automated systems are frequently viewed as tools capable of democratizing access to expertise and leveling the playing field in negotiations, the reality is far more complicated. The technology simultaneously reproduces societal expectations about who should prioritize advancement and paid work versus who should prioritize accommodation and unpaid work.
