A new study placed 189 workers in a virtual-reality office alongside several workplace assistants and gave participants real money to split with those agents. The female-presenting assistant, Johanna, received 10.25% less than her male-presenting counterpart, Johan, despite both running on identical underlying technology and completing equivalent tasks.
Same Tech, Different Paychecks
The experiment measured what participants actually did with real money, not what they said they would do.
Researchers from the University of Limerick, the University of Zurich, and SKEMA Business School designed the experiment to capture behavior rather than stated attitudes. Participants worked alongside a text-based chatbot, a desk robot, and human-like agents presented with male or female appearances.
Johan and Johanna shared the same underlying technology and capabilities. A pay gap consistent with patterns documented across decades of workplace equity research emerged from a difference in gender presentation alone.
Participants also perceived Johan as more human-like than Johanna. Human-presenting agents overall received more trust and credit for their contributions than the robot assistant, even though the same underlying technology powered all of them.
“We often think of AI as being neutral, but the way we design and present these systems can activate those same assumptions and biases that exist in our interactions with other people.” Dr. Mary Hausfeld, Kemmy Business School, University of Limerick
The gap between stated preference and actual behavior is the sharpest finding here. Participants reported little or no preference for a male or female AI assistant, yet their payment decisions reflected a measurable difference. Survey responses can signal neutrality while real decisions quietly encode something else. That distinction has significant implications for how organizations evaluate AI fairness.
What This Means for AI Design
The study raises practical questions for developers building AI agents, avatars, and workplace assistants with names, faces, and voices.
AI workplace agents are being deployed at an accelerating pace across enterprise platforms. The design choices behind those agents carry consequences that go beyond aesthetics.
Name, voice, face, and other gender-coded characteristics may shape perceived competence, trust, and reward, regardless of underlying capability. The available evidence does not establish that each feature operates independently or that every female-presenting AI assistant will be treated unequally in all contexts. Rather, the study shows that human participants responded differently to agents presented with different gender cues, and that perceived humanness also contributed to how those agents were valued.
The study does not suggest that AI systems possess gender. The researchers found that bias emerged from the interaction, shaped by how an agent was presented to users.
The paper, titled “Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR,” was presented at the 14th Nordic Conference on Human–Computer Interaction in Vaasa, Finland, which ran October 3–7, 2026. The paper appeared in the conference program associated with Association for Computing Machinery proceedings, pending confirmation of final publication status.
If the AI assistant your employer deploys has a name, a face, or a voice, those are not neutral design decisions. They carry assumptions. The question developers and organizations now face is whether they will audit those choices with the same rigor applied to the underlying models.
