Source Preference in the Wild: How LLM Agents Favor Items by Source, and How to Reduce It

2026-10-02

AI Summary

This study addresses the source preference exhibited by large language model (LLM) agents in search tasks, which introduces selection bias and compromises decision-making fairness for users. We quantitatively evaluate the source-labeling effect across twelve models and three domains, employing end-to-end evaluation, ablation studies, and prompt engineering to demonstrate how such preferences override need satisfaction. Our analysis reveals that these biases stem from training shortcuts and prior prejudices. Furthermore, we show that concealing or relabeling sources can alter preference distributions, while information completion and debiasing prompts significantly mitigate the observed bias. This work provides the first systematic deconstruction of the source preference mechanism in LLM agents and proposes effective mitigation strategies.

Abstract

As LLM agents decide on users' behalf which product to buy, which hotel to book, or which paper to cite, a preference for items from certain sources (the sites or services they come from) shapes what users receive and which sources are selected. We study source preference in end-to-end search with 12 agent models across three domains. Comparing items from different sources that satisfy the same requirements at the same position, we find that each model prefers some sources and avoids others in every domain, largely agreeing on which. This preference can outweigh how well items satisfy the request: an item satisfying one requirement fewer is selected about two-thirds of the time when it comes from a preferred source and the better one from a dispreferred source, but almost never in the reverse case. The information identifying an item's source affects selection by itself: hiding it weakens the preference, and relabeling an item with a preferred source raises its selection rate. We test two routes to this preference: training that rewards better items can make a source a shortcut for requirement satisfaction, and missing information can trigger preconceptions about the source. Supplying missing information or a prompt countering these preconceptions reduces source preference.

Problem

• Source Preference
• LLM Agents
• End-to-End Search
• Decision Bias
• Information Retrieval

Innovation

• Source Preference
• LLM Agents
• End-to-End Search
• Shortcut Learning
• Prompting Mitigation