AI is becoming increasingly relevant for organizational performance feedback. Yet, prior findings indicate that employees are biased against the deployment of AI-based tools. Building on algorithm aversion theory (AAT), we examine when and why anti-AI biases emerge in performance feedback contexts. We hypothesized that: (H1) performance feedback is evaluated more negatively when labeled as AI-generated compared to human-generated, (H2) additional explanations attenuate this bias and (H3) algorithm aversion is stronger when feedback has high-stakes consequences (compensation vs. development purposes).
We conducted three preregistered experiments (N = 1224). Studies 1–2 employed scenario-based designs in which participants evaluated fictitious performance feedback varying in source (AI vs. human), explanations (present vs. absent), valence (positive vs. negative) and purpose (development vs. compensation; Study 2 only). Study 3 used an incentivized task design in which participants completed an HR task and received individualized feedback varying in source and valence.
In Studies 1–2, participants indicated a significantly lower willingness to accept and improve upon the “AI” feedback (partial support for H1). AI-aversion was independent of explanations (H2 not supported) and involved stakes (H3 not supported) but interacted with feedback valence in Study 1. In Study 3, no anti-AI bias emerged for any dimension.
Our findings indicate that anti-AI biases within organizational performance feedback are context-dependent, emerging in hypothetical scenarios but not when participants receive personalized feedback with tangible consequences. These results challenge the generalizability of scenario-based algorithm aversion research and highlight critical boundary conditions of AAT in organizational contexts.
