Researchers found that major AI models responded differently to workplace prompts depending on subtle gender-associated language patterns. Across GPT-4, Llama, Gemma and Mistral, prompts containing language more commonly associated with women produced less formal and less complex professional writing than male-associated prompts, even when the user’s name did not signal gender. The findings suggest that linguistic style alone can trigger biased differences in AI-generated workplace communication.

https://arxiv.org/abs/2608.13328