Johns Hopkins Researchers Identified AI Gender Bias

A study found that AI models generate more emotional and less formal text when responding to female-coded language.

Updated on Sept. 28, 2026 in Artificial Intelligence

Isometric editorial illustration of silicon wafers and geometric architectural volumes, representing structural bias in artificial intelligence models. — Researchers at Johns Hopkins University identified that AI models generate different levels of formality and emotion based on gender-coded inputs in prompts. AI Illustration. Upload story photo >

Live Poll

Should AI writing tools remain neutral rather than mimicking the tone of user prompts?

Researchers at Johns Hopkins University discovered that popular AI models adopt different tones based on the gendered nature of prompts. The study revealed that AI responses become less formal and more emotional when they encounter language coded as female.

Why it matters

Understanding how AI models interpret and mirror gendered language is crucial for ensuring equitable and professional automated communication. These patterns highlight how AI biases can inadvertently shift the tone of workplace interactions.

Researchers tested four prominent AI models, including GPT-4, Llama, Gemini, and Mistral, to analyze word choice and tone. The team observed distinct shifts in style when input prompts contained female-coded terms like maybe, we, and wonderful.

The players

Data Science and AI Institute at Johns Hopkins University

This academic center conducts interdisciplinary research into the development and societal implications of artificial intelligence technologies.

The details

The study from the Data Science and AI Institute at Johns Hopkins University involved running workplace prompts through various models to compare generated email responses. While female-coded inputs triggered emotional or informal language, substituting names such as John within those same prompts did not alter the resulting response style.

September 28, 2026: The research findings were published.

The Big Picture

This discovery shifts the understanding of AI development by highlighting how latent gender biases persist in large language models. It challenges the assumption that AI remains neutral, forcing researchers to re-examine how models bridge the gap between human language and automated output.

Users relying on AI for professional correspondence should review generated outputs for unexpected changes in tone or emotional framing. Being aware of these biases allows for better manual refinement of AI-written emails to ensure they maintain the intended professional standard.

The takeaway

AI models currently lack the objective neutrality many users assume, often mirroring specific gendered patterns in their word choice. Users should maintain a critical eye when using these tools for professional communication until developers implement more robust bias mitigations.