Contaminated Collaboration: Measuring Gender Bias Transfer in LLM-Assisted Student Writing

2026-06-14

AI Summary

This study investigates whether gender bias embedded in large language models (LLMs) can be transmitted to student writing through AI-powered writing assistants. Through a controlled experiment, the authors compare career-planning essays produced by students under three conditions: no AI assistance, neutral prompts, and gender-biased prompts. Combining linguistic feature analysis with statistical testing, the study provides the first empirical evidence that LLM-induced bias can transfer to human-authored texts, with an asymmetric effect: agency in texts related to women is significantly diminished, whereas texts about men remain largely unaffected. The findings demonstrate that biased prompts exacerbate disparities in perceived agency and reinforce gender-stereotypical career recommendations, highlighting substantial risks of bias propagation in human-AI collaborative writing environments.

Abstract

Gender bias in LLMs has been studied extensively in model outputs, with biased prompts shown to amplify stereotyped generations. Whether such bias propagates into text produced by humans who use these systems, however, remains underexplored. We investigate whether gender bias in an LLM writing assistant transfers into career plan essays written by students. We first verify that a gender-biased prompt induces gender-differentiated language in LLM-generated essays, while a neutral prompt does not. We then recruited participants (N = 123) in a controlled environment to write career plan essays for paired biographical profiles differing only in gender under three conditions: no AI assistance, neutral LLM assistance, or gender-biased LLM assistance. Students in the biased condition produced essays with a significantly larger agentic gap and more gender-stereotypic occupation suggestions than those in the control and neutral conditions. Our results also reveal that this bias transfer is asymmetric: agency is suppressed in female-target essays while male-target writing remains largely unaffected. Our findings highlight the risk of bias propagation in AI-assisted writing, calling for fairness-aware design in educational AI tools.

Problem

• gender bias
• bias transfer
• LLM-assisted writing
• stereotyping
• educational AI

Innovation

• bias transfer
• gender bias
• LLM-assisted writing
• agentic gap
• fairness in AI