Who Gets Cited? Gender- and Majority-Bias in LLM-Driven Reference Selection
2025/08/02 by Jiangen He, He, Jiangen · 17 voices · 1 citation
#cs.DL #cs.AI #cs.CY
paper · pdf · doi:10.48550/arxiv.2508.02740
Abstract
Large language models (LLMs) are rapidly being adopted as research assistants, particularly for literature review and reference recommendation, yet little is known about whether they introduce demographic bias into citation workflows. This study systematically investigates gender bias in LLM-driven reference selection using controlled experiments with pseudonymous author names. We evaluate several LLMs (GPT-4o, GPT-4o-mini, Claude Sonnet, and Claude Haiku) by varying gender composition within candidate reference pools and analyzing selection patterns across fields. Our results reveal two forms of bias: a persistent preference for male-authored references and a majority-group bias that favors whichever gender is more prevalent in the candidate pool. These biases are amplified in larger candidate pools and only modestly attenuated by prompt-based mitigation strategies. Field-level analysis indicates that bias magnitude varies across scientific domains, with social sciences showing the least bias. Our findings indicate that LLMs can reinforce or exacerbate existing gender imbalances in scholarly recognition. Effective mitigation strategies are needed to avoid perpetuating existing gender disparities in scientific citation practices before integrating LLMs into high-stakes academic workflows.
Cited by
Discussions
- Here is work showing some of this effect. It is important, however, to see the form that this takes ‘in the wild’ in terms of the erasure of actual women-led scholarship in LLMs by multiple modalities [bsky, 544 points, 5 comments]
- Don’t trust LLMs with your literature searches. Horrible gender biases as found in this paper (from later in the thread below) arxiv.org/abs/2508.02740 🧪🔭 [bsky, 84 points, 1 comments]
- Thanks! Link: arxiv.org/abs/2508.02740 [bsky, 73 points, 0 comments]
- I had the sane question (I've got a paper under review making similar points). In the replies to another part of the thread, this came up: arxiv.org/abs/2508.02740 I've not checked it out yet, but mig [bsky, 21 points, 1 comments]
- Someone else posted their own study here. I'd also love to see more. - arxiv.org/abs/2508.02740 [bsky, 20 points, 0 comments]
- Heres the sourse for the info in the post quoted below… arxiv.org/abs/2508.02740 [bsky, 11 points, 2 comments]
- Cool, thank you. For those playing along at home, here's the link: arxiv.org/pdf/2508.02740 [bsky, 6 points, 1 comments]
- Related: arxiv.org/abs/2508.02740 [bsky, 4 points, 2 comments]
- * AI emphasizes mysoginy, erasing female contributions and/or replacing them with hallucinated male ones (arxiv.org/abs/2508.02740) * LLM dissolves social interactions in favir of individual ones. Par [bsky, 1 points, 1 comments]
- preprint: arxiv.org/pdf/2508.02740 [bsky, 1 points, 1 comments]
- "Överraskande" att AI-språkmodeller bidrar till osynliggörande av kvinnor i akademin. arxiv.org/abs/2508.02740 [bsky, 1 points, 0 comments]
- The paper for this: arxiv.org/abs/2508.02740 [bsky, 0 points, 0 comments]
- En el tema de la progresión en la escala científica, el impacto de las publicaciones se mide, entre otras cosas, en citas. Pues bien, se sabe que hay un sesgo claro a citar menos las publicaciones de [bsky, 0 points, 1 comments]
- > “AI” is useful…. for misogyny: https://bsky.app/profile/bayesianboy.bsky.social/post/3moh5k37gzk2s https://arxiv.org/abs/2508.02740 [bsky, 0 points, 0 comments]
- @rhyall.bsky.social this is why y’all bot didn’t find Sarah when you asked to show her feed. The fascist owners of the AI companies are deliberately de-algorithming the contributions of women and mino [bsky, 0 points, 0 comments]
- Actually the Matilda Effect is one of the most robustly evidenced results in social science. also: arxiv.org/abs/2508.02740 [bsky, 0 points, 2 comments]
- arxiv.org/abs/2508.02740 [bsky, 0 points, 0 comments]
Related