When Wording Steers the Evaluation: Framing Bias in LLM judges

TL;DR

This paper investigates how subtle prompt phrasing, inspired by psychological framing effects, biases large language model judgments across multiple high-stakes evaluation tasks, revealing a structural vulnerability in current LLM evaluation methods.

Contribution

It systematically demonstrates that prompt framing significantly influences LLM judgments, highlighting the need for framing-aware evaluation protocols.

Findings

Framing induces significant discrepancies in model outputs.

Different LLM families show distinct framing tendencies.

Framing bias is a structural property of current LLM evaluation systems.

Abstract

Large language models (LLMs) are known to produce varying responses depending on prompt phrasing, indicating that subtle guidance in phrasing can steer their answers. However, the impact of this framing bias on LLM-based evaluation, where models are expected to make stable and impartial judgments, remains largely underexplored. Drawing inspiration from the framing effect in psychology, we systematically investigate how deliberate prompt framing skews model judgments across four high-stakes evaluation tasks. We design symmetric prompts using predicate-positive and predicate-negative constructions and demonstrate that such framing induces significant discrepancies in model outputs. Across 14 LLM judges, we observe clear susceptibility to framing, with model families showing distinct tendencies toward agreement or rejection. These findings suggest that framing bias is a structural property of current LLM-based evaluation systems, underscoring the need for framing-aware protocols.