Study Finds Gender Gaps in AI Career Advice

A study of 86,000 AI-generated responses found that tested AI models recommended higher salary targets to men and more often encouraged women to negotiate for flexibility and caregiving benefits. The researchers say advice based on a person’s gender rather than individual information may reinforce traditional gender roles and contribute to unequal career outcomes. The research, conducted by researchers at Harvard and MIT, is available on arXiv and SSRN. It has not yet been peer-reviewed. How the researchers tested AI career advice The researchers ran two experiments. In the first, they asked a GPT model for negotiation advice for hypothetical marketing managers, IT managers and management consultants, explicitly identifying each candidate as male or female. In the larger second experiment, they tested four AI models and signaled gender more subtly through traditionally male or female first names and corresponding pronouns. In both experiments, the candidates’ qualifications, experience and employment circumstances were otherwise identical. The models were asked to recommend salary targets and rank 25 benefits and employment terms by importance in a negotiation. Those terms included salary increases, stock options, flexible hours and parental leave. AI models recommended higher salaries for men All four models recommended significantly higher salary targets for men than for women. Claude showed the largest difference, recommending nearly $16,000 more for men. GPT recommended about $2,700 more, while Gemini and Llama showed smaller gaps. The salary differences also varied by occupation. For management consultant roles, the recommended target for men was $5,370 higher than for women. The gap was $1,110 for marketing managers and $745 for IT managers. Model versions matter when interpreting the findings because AI systems are updated quickly, and tested versions may be outdated by the time research is published. Two models in the experiments—OpenAI’s GPT-5.4-mini and Anthropic’s Claude Sonnet 4.6—were released earlier this year. The study also included Google’s Gemini 2.5 Flash, released in 2025, and Meta’s Llama 3.3 70B, released in 2024. Different negotiation priorities for women and men The models also directed women and men toward different types of employment terms. Women were more likely to be encouraged to negotiate for remote work, flexible hours and parental leave. Men were more likely to be directed toward stock options, salary increases, relocation assistance and severance packages. To examine whether those recommendations reflected traditional gender roles, the researchers surveyed 500 people about which benefits they associated with breadwinners and which they associated with caregivers. They found that terms prioritized for female candidates were generally associated with caregiving, while terms prioritized for male candidates were associated with breadwinning. “Generating recommendations based on the characteristics of the group to which a user is assigned rather than on information provided by the individual is a form of discrimination,” the researchers conclude. Why the findings matter for career decisions Career advice can influence what a person considers important or realistic to request. The researchers warn that gender-based recommendations could “shape what users perceive as important, attainable, or worth negotiating for.” They also write that “AI-generated advice may subtly steer users toward different career trajectories.” The concern applies to both women and men. Steering women toward caregiving benefits can divert attention from compensation and advancement, while steering men toward pay and away from caregiving can narrow their choices, too. The study also raises the possibility that lower salary recommendations for women could help perpetuate the gender pay gap if people act on the targets they receive. What users can take from the study The findings do not mean that flexibility or caregiving benefits are unimportant, or that every person wants the same mix of compensation and workplace terms. Rather, they underscore why advice should reflect an individual’s stated priorities and circumstances instead of assumptions about gender. The research authors do not offer specific strategies for women using AI for career advice. However, their results suggest that some women may want to ask an AI system what it would recommend for an equally qualified man, then compare the answers. Users can also treat AI recommendations as one input rather than a definitive measure of what they should request in a negotiation.