Job seekers increasingly use (LLMs), a type of artificial intelligence that generates text, to write cover letters, and employers use similar AI tools to screen them. This creates a hidden bias: the LLM used to write a letter can affect screening outcomes, regardless of the applicant's true qualifications.
Researchers tested this with eleven LLM writers and nine LLM evaluators across ten job descriptions. They fixed candidate quality and varied only the writer model to isolate the effect.
They found that evaluators favor certain writers and even prefer their own outputs. In the worst case, about two out of twenty-five qualified candidates could be filtered out due to model mismatch alone. This reveals a new source of hiring inequity that regulators should address.