Designed to Please: How Sycophantic AI Voices Reinforce Gendered Expectations

Scholars have raised concerns that the popularization of feminized, agreeable AI assistants normalizes associations between women and subservient behavior, potentially reinforcing broader gender stereotypes through repeated exposure. This study tested whether AI sycophancy and AI gender shape user perceptions of AI warmth, competence, affective trust, and cognitive trust, and whether interacting with a sycophantic female AI shifts explicit or implicit gender stereotype endorsement. Using a 2 (AI gender: male vs. female) × 2 (sycophancy: high vs. low) between-subjects design, 202 participants completed a simulated travel planning task with a voice-based AI agent. Despite successfully manipulating sycophancy, neither sycophancy nor AI gender impacted gender stereotype endorsement. In fact, sycophancy produced no significant effects on any outcome. The only significant main effect of AI gender was on warmth, where male-coded AI was rated warmer than female-coded AI, contrary to predictions. The study's most robust finding was an in-group favorability pattern: male and female participants had greater affective trust in AI matching their own gender, and male participants also showed additional in-group favorability effects on ratings of warmth and cognitive trust. The null stereotype results do not resolve concerns about feminized sycophantic AI. Rather, they suggest that a single brief interaction is unlikely to reinforce gender stereotypes. The in-group favorability findings further challenge the assumption that a female-coded AI assistant functions as a neutral universal default.