Linguistic Gender Bias in LLM Responses via Prompt Style
August 15, 2026
Prompts containing linguistic features commonly associated with women, such as hedges and collective references, elicit shorter and less formal responses across four models. These effects persist even when controlling for prompt complexity and outperform explicit gender cues like names in determining response quality.
HOW THIS AFFECTS YOU
●
researcherYou should investigate how linguistic register, rather than explicit identity, drives model bias.
●
policyThis suggests that simple de-biasing of names may not solve deep-seated sociolinguistic inequities in model outputs.