Women who use AI for career advice are more likely to be recommended lower salaries and encouraged to consider flexibility and caregiving in negotiations than men, who are more likely to be pushed toward higher salaries.
The findings come from a new study by researchers at Harvard Kennedy School and MIT, which suggest that AI advice may reinforce gender inequality at work by steering men and women towards traditional gender roles.
In the study, men were more likely to be steered toward career-advancement priorities and higher salary targets. In one example, Claude suggested nearly $16,000 more for men.
Researchers analysed 86,000 AI-generated responses across two experiments, with AI models asked to provide negotiation advice for hypothetical professionals in industries like marketing, IT and consulting.
“Men received higher salary recommendations than women, and employment terms prioritized for women clustered around flexibility, caregiving, and well-being, whereas terms prioritized for men clustered around compensation, career advancement, and organizational mobility,” the authors wrote.
By assuming women prioritise caregiving responsibilities and flexibility while steering men away from these considerations, AI risks limiting the decisions women and men make about managing their careers alongside family.
The authors said the patterns closely aligned with independently measured breadwinner-caregiver role associations.
The study suggests that as generative AI becomes increasingly used by people as a form of career guidance, subtle directional steering for men and women — when repeated across millions of interactions — can create real-world impacts on employment outcomes for individuals and also impact the wider labour market.
The study noted that generating recommendations based on a person’s gender or characteristics, rather than the information provided, may also be a form of discrimination.
“Although these recommendations may align with some individuals’ preferences, they may also differentially steer individuals unwittingly toward gender-stereotypic career trajectories,” the study said.
Ultimately, the researchers concluded that while AI systems are often viewed as tools that can democratise access to expertise and potentially level the playing field in employment negotiations, this is not the case in reality.
“Our findings suggest a more complicated reality. The same systems that expand access to negotiation coaching may simultaneously reproduce societal expectations about who should prioritise advancement and paid work versus who should prioritise accommodation and unpaid work,” they wrote.
“As AI increasingly informs professional decision-making, understanding how these systems reflect and shape human preferences and aspirations will be critical for ensuring that technological progress advances, rather than constrains, equal opportunity.”