LLM-based reframing of news headlines increases conservative readers' trust in liberal media, but models overestimate their own effectiveness

Two pre-registered experiments tested whether LLM-generated debiasing could improve cross-partisan receptivity to news. Study 1 found that subtle lexical debiasing (replacing emotive words with moderate synonyms) had no effect on human readers. Study 2 found that substantive reframing significantly increased conservatives' perceived trustworthiness, completeness, and willingness to engage with liberal news headlines, without producing a backfire effect among liberals.

LLM-simulated participants showed robust effects in Study 1 where human readers did not; in Study 2, the model's predicted effects aligned directionally with human responses but were significantly larger in magnitude for some outcomes. The divergence suggests current LLMs lack the quantitative accuracy and qualitative psychological fidelity to evaluate their own interventions.

The findings demonstrate that LLM-based debiasing targeting ideological framing rather than surface-level language can improve cross-partisan receptivity, but current models require human oversight to avoid overestimating intervention effectiveness.