Job candidates Emily Clarke and James Clarke had the same résumé. The exact same.
Evaluators knew both applicants had used AI, but when reviewing the résumés, they were twice as likely to question Emily's competence.
Zehra Chatoo, founder of Code For Good Now, calls this the "AI judgement penalty." This term describes the double-bind women experience when they are asked to use AI but then are penalized for doing so.
"I was really trying to understand the fear of judgement women have around using AI. Is this grounded in reality? Is it truth or perception? So, we decided to put it to the test," Chatoo explained.
The experiment was simple. Chatoo created one résumé for the same job posting, then swapped the first name before handing it to people in her study. The only difference was the name.
The AI usage was the same. The work was the same. The person was the same.
But the advantage wasn't.
When we think about equity and AI, much of the conversation has focused on access and upskilling. These gaps still need to be closed. But Emily Clarke's resume brings to light a new question: who is penalized and who is rewarded for using AI?
Seventy-three percent of women say they use AI at least weekly according to a Catalyst survey of more than 1,500 women across the US and Canada.
Yet 45% of women feel caught between using AI to perform and avoiding it to protect their credibility.
They all share the same challenge as Emily Clarke.
There is a growing body of research showing that AI, implemented responsibly, can be an equalizer, closing gaps in performance, and reducing disadvantages.
A global study of developers by LeanIn found that 64% of women believed AI accelerated their careers. They were learning faster, solving problems more efficiently, and contributing with greater confidence.
Other studies have shown AI as effective in reducing education and language gaps. A randomized study by the National Bureau of Economic Research showed that generative AI closed the initial performance gap between higher and lower levels of education, by three-quarters.
A team of information science researchers from the University of Wisconsin–Madison, Peking University, and the University of Hong Kong found that among the biomedical scientific community, AI-assisted writing grew twice as fast among non-English speaking countries compared to English speaking countries.
Similarly, preliminary research has demonstrated that neurodivergent employees benefit from AI tools that are used to enhance communication and reduce sensory overload.
These studies demonstrate the positive potential of AI, but the promise of AI isn't guaranteed. A group of AI ethics and information science researchers from the University of Washington and Indiana University found that AI can proliferate bias. Microsoft researchers have found that AI usage can degrade critical thinking.
Access tells us who has the tool. It doesn't tell us who can use it to overcome a barrier, who receives credit for using it, or who benefits when productivity increases. That is the AI advantage gap.
Women provide a view into the access vs. advantage problem, but they are not the only group where access doesn't always equal advantage.
Only about 38% of frontline workers report using AI according to a global UKG study. While 43% are optimistic about using AI, two-thirds are concerned that AI will replace them.
The concern is partially driven by how frontline workers experience AI. While an office worker might be encouraged to experiment with AI, frontline employees have less influence over its implementation and fewer opportunities to develop the skills needed to benefit from it.
For many frontline workers, AI may already be embedded in the systems, tools, and workflows they use. Yet they aren't always included in decisions about how those technologies are introduced. Research from PwC and the Manufacturing Institute found that 45% of manufacturing leaders identified excluding frontline leaders from AI design and rollout as a significant contributor to unsuccessful AI initiatives.
Forthcoming Catalyst research with Charta der Vielfalt adds another dimension, identifying a 15 percentage-point gap between management and non-management employees in their experience of the transparency, training, and voice needed to navigate technological change.
Here, measuring only the adoption rate hides a larger question: Who has the opportunity to shape how AI changes their work, develop new capabilities, and benefit from the results?
For frontline workers, the advantage isn't simply using AI. It's having opportunities to learn, influence how the technology is applied, and contribute in new ways.
Giving someone access to powerful technology is only one step in a much longer chain:
Access → Opportunity → Capability → Application → Credit → Advantage
Inequality can enter at any point in this chain depending on who has the time to learn and experiment, and whose work is trusted and receives credit.
Zehra Chatoo believes AI can be a “great leveler.” But, she says, it must be “pointed in the right direction.”
Getting there will require organizations to move beyond a skills-first approach.
Chatoo calls training "table stakes" and invites leaders to take it a step further. Start by creating accountability. Try to understand who is benefiting and where the advantage lies. Then start building a culture where employees can use AI without facing the double-bind.
This is essential because the people who use AI, shape it.
"AI is a technology that is shaped by its user base. And so, if less women use it, that means less women's voices will be represented in the output," says Chatoo.
Measure what happens after access.
Go beyond adoption rates. Start measuring employee skill sets, critical thinking, employee’s understanding of AI, its use cases and its limitations.
Create permission to experiment.
Create a culture where there are people at every level modeling proper usage. Encourage safe experimentation and support employees that use AI appropriately.
Teach judgment, not just prompting.
Help employees go deeper than prompting. Teach them when to trust an output, when to challenge it, and when to report an issue.
Look for where people fall off the path to advantage.
Understand where in the chain people are experiencing inequities. Some employees may have limited access, training, or agency. For women, it could be credibility.
AI has enormous potential to close education, accessibility, and even language barrier gaps. But implementing AI doesn't guarantee those outcomes. What AI becomes is determined by who has access, who feels safe using it, and whose voice is included.
The next AI divide won't be about access, it will be about who owns the advantage. Organizations that win the AI race, won't just have more people using AI; they'll have more people shaping what AI means for their workforce.
Chatoo, Z. (May 14, 2026). The AI judgment penalty. Creativebrief.
Shaffer, E., Gutierrez, L., Brassel, S., & Henneborn, L. (2026). AI’s hidden costs for women: Risks to talent and ROI. Catalyst.
Noble-Tolla, M. (July 2, 2026). How the AI Gender Gap Will Compound the Leadership Gap. Lean In.
Cruces, G., Fernández Meijide, D., Galiani, S., Gálvez, R. H., & Lombardi, M. (2026). Does Generative AI Narrow Education-Based Productivity Gaps? Evidence from a Randomized Experiment. National Bureau of Economic Research, Working Paper 34851.
Liu, J., He, Y., Zheng, Z., Bu, Y., & Ni, C. AI-Assisted Writing Is Growing Fastest Among Non-English-Speaking and Less Established Scientists.
George, G., Kulkarni, M., & Varghese, B. (2026). AI in creating inclusive work environments for neurodiverse employees. Advances in Autism, 12(1), 79–98.
UKG. (2026). AI and the Frontline Workforce.
PwC & the Manufacturing Institute. (March 31, 2026). Frontline leadership in manufacturing’s AI adoption.
Wilson, K., Sim, M., Gueorguieva, A.-M., & Caliskan, A. (2025). No Thoughts Just AI: Biased LLM Hiring Recommendations Alter Human Decision Making and Limit Human Autonomy. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(3), 2692–2704.
Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. Microsoft Research and Carnegie Mellon University.