Evaluating demographic framing mitigation in LLM rewriting
November 20, 2026
An adaptive rewriting policy that proposes edits on both sides of a demographic contrast outperforms fixed policies targeting low persuasion bias. Testing across 660 gender and 1,980 age pairs shows this approach improves consistency in mitigating framing disparities.
HOW THIS AFFECTS YOU
●
researcherAdaptive policies are more effective than fixed rules for reducing demographic bias in text.
●
policyThis provides a way to mitigate bias in LLM outputs when the original model is inaccessible.