AI-powered recruitment systems may promise faster, more objective and unbiased hiring, but new research suggests that applicants’ perceptions of fairness are shaped as much by the appearance of AI interview avatars as by the hiring decision itself, revealing an unexpected psychological challenge in AI-driven recruitment.
A study by Ka Hei Carrie Lau and colleagues, published in Skin-Deep Bias: How Avatar Appearances Shape Perceptions of AI Hiring in the Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (2026), found that job seekers perceived AI-driven rejection as least fair when the interview avatar shared only one characteristic with them, either gender or skin color.
Artificial intelligence is rapidly transforming recruitment processes worldwide, with many companies relying on AI not only to screen CVs but also to conduct job interviews through human-like digital avatars. Businesses are increasingly adopting these systems to speed up hiring while reducing human bias in recruitment.
However, new research indicates that while AI may be designed to make impartial decisions, applicants often judge the fairness of those decisions based on the avatar’s appearance, highlighting an overlooked psychological factor in AI-driven hiring.
Human-like AI creates social responses
Researchers from the Technical University of Munich (TUM) and Lund University examined how applicants perceive AI interview decisions depending on the visual characteristics of recruitment avatars.
The study involved around 220 participants from Germany, the United Kingdom and the United States. Each participant completed a simulated job interview for a fictional customer support position, interacting with a photorealistic AI avatar capable of asking follow-up questions and responding in a human-like manner.
To assess whether appearance influenced perceptions, researchers created four versions of the AI interviewer. The avatars varied by gender and skin color, appearing as either male or female and with either light or dark skin.
Participants’ eye movements were monitored using eye-tracking technology throughout the interviews before they completed detailed questionnaires. The findings were published in the Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems.
Eye tracking reveals visual attention patterns
The eye-tracking analysis found that participants spent more time looking at an avatar’s face when its skin color differed from their own, based on self-reported information.
Despite these differences in visual attention, participants generally expressed high levels of trust in the AI interviewer regardless of whether the avatar matched their gender or skin color. The perception of fairness shifted significantly after every participant received the same outcome, a rejection for the fictional job.
Shared traits influenced perceptions of fairness
Following the rejection, applicants became more likely to believe they had not been evaluated impartially.
Researchers found that participants whose skin color differed from the avatar were more likely to attribute the rejection to bias. However, the strongest negative reaction came from applicants who shared only one characteristic with the AI interviewer, either gender or skin color, but not both.
This group rated the recruitment decision as less fair than participants who matched the avatar in both characteristics. They also perceived the decision as more unfair than participants who shared no characteristics with the avatar at all.
The findings suggest that partial similarity between applicants and AI interviewers may create stronger expectations of fairness, making rejection feel more personal and less objective.
Study highlights new challenge for AI recruitment
The researchers concluded that discussions surrounding fairness in AI recruitment should extend beyond eliminating bias in algorithms and training data.
The study found that even when AI systems are designed to operate impartially, applicants may still perceive decisions as unfair because of subconscious social reactions triggered by the appearance of human-like avatars.
According to the research, understanding how people socially respond to AI interviewers will be essential for designing recruitment technologies that are widely trusted and accepted. The findings highlight the need for AI developers and employers to consider both technical fairness and human psychology as AI-driven recruitment becomes more common across industries.
Also Read | Warning: Agentic AI browsers may put user data at serious risks!
