Whose startup does the AI back? An algorithm audit of reputation and demographic signals in LLM-assisted crowdfunding
| Source: arXiv AI
Tags: algorithmic audit, conjoint experiment, crowdfunding, LLM bias, Journal of Business Venturing Insights
An audit of seven LLMs advising crowdfunding backers found founder gender and race barely matter by default, but models favor Black- and female-signaled founders when users say they want to back underrepresented creators.
Details
LLM assistants increasingly advise backers on which crowdfunding campaigns to support. This study asks when and how they use founders' gender and ethnicity, using a pre-specified conjoint experiment where seven LLMs chose among five synthetic campaigns with randomized quality signals and founder names, across three backer personas and a control, for 40,248 responses. By default founder names barely mattered, and under budget-conscious and delivery-focused personas the gender and Black-name effects were statistically equivalent to zero. When the backer wanted to support underrepresented creators, models favored Black-signaled founders by 8.1 points, positive in all seven, and female-signaled founders by 5.1 points, mostly among near-tied campaigns. The models almost never cited gender or ethnicity in their reasons. The authors read this as the assistant taking demographic preferences from its user rather than its training data.