Understanding Gender Bias in AI-Generated Product Descriptions

Gender bias in e-commerce AI is a blind spot: While gender bias in large language models has been studied broadly, AI-generated product descriptions for e-commerce have received little attention despite being widely deployed, meaning potential harms to consumers may go undetected and unaddressed.

AI makes problematic assumptions about clothing and body types: The research identified a novel form of bias where AI systems make stereotypical assumptions about clothing sizes based on gender, suggesting that e-commerce AI can perpetuate harmful body image standards in ways that general-purpose bias detection may miss.

Different products get different sales pitches based on gender: The study found that AI systems use stereotypical bias when deciding which product features to highlight, meaning identical or similar items may be advertised differently to different genders—potentially influencing purchasing decisions unfairly.

Persuasive language varies by gender in AI descriptions: The research discovered that AI-generated descriptions use different levels and types of persuasive language depending on the gender context, which could systematically advantage or disadvantage certain products based on how they're gendered.

Current AI harm frameworks need e-commerce-specific approaches: The findings demonstrate that existing taxonomies for AI harms (exclusionary norms, stereotyping, and performance disparities) apply to e-commerce but require specialized detection and mitigation methods tailored to this specific commercial context.