Beyond the Default: How Customizable Artificial Intelligence Agents Can Attenuate Stereotypical Preferences

Abstract

AI agents are typically deployed with gender cues that match traditional role expectations, such as female voices or avatars in assistive and support roles, because stereotype‐congruent designs enhance consumer adoption, trust, and engagement. However, prior research suggests that this practice risks entrenching occupational gender stereotypes at a societal level, raising the question of whether, and under what conditions, consumers’ stereotype‐congruent preferences can be attenuated. Drawing on Role Congruity Theory and research on bias correction, we propose that preferences for gender‐stereotypical AI agents are not fixed but contingent on the salience of gender cues in the decision context: when gender is foregrounded as a potential source of bias, consumers may deliberately shift toward counterstereotypical choices. Four experimental studies ( N  = 2530) test this proposition. Study 1 establishes that, under low gender‐cue salience, consumers default to AI agents whose gender aligns with stereotypical occupational roles. Study 2 shows that making gender cues salient in the choice process can attenuate these default preferences, and, under certain conditions, shift choices toward counterstereotypical agents. Studies 3 and 4 provide process evidence consistent with a role for action‐efficacy beliefs, whereby individuals come to view choosing a counterstereotypical AI agent as a meaningful way to promote gender equality. Gender salience, whether induced through design choices (Study 3) or explicit bias warnings (Study 4), was associated with stronger action‐efficacy beliefs, and the conditional indirect effects through these beliefs emerged primarily among women in male‐typed contexts, while comparable effects were weaker or absent among men and in female‐typed domains. Our findings identify conditions under which AI design choices can attenuate stereotype‐congruent preferences and provide evidence regarding the psychological processes associated with these shifts, offering actionable insights for managers and policymakers seeking to design AI agents that support more equitable human–AI interactions.