Positional bias affects recommendations from large language models
Statistical Detection of Positional Bias in LLM-Based Recommender Systems: Methodologies and Empirical Evidence
Information RetrievalArtificial Intelligence
Summary
This paper finds that big language models used in recommendation systems tend to favor items based on where those items appear in their input list, rather than just their quality. The researchers developed two methods to detect this positional bias and tested them using a real product catalog with many user queries. Their results show that this bias is a real, measurable effect that depends on the model and setup, not just how the recommendations are generated. This means recommendations could be unfair or less accurate simply because of item order, so routine checks and controls are suggested.
What this means in practice
- •For conversational ai developers: Detect and mitigate positional bias in prompts to improve fairness and relevance of AI-driven product recommendations.
- •For ecommerce platform teams: Implement auditing and governance controls such as order randomization to ensure unbiased item rankings from LLM-based recommenders.
Authors
Carlos Peláez-González, Francisco Herrera, Andrés Herrera-Poyatos
Abstract
Large Language Models (LLMs) are increasingly used as recommendation generators in conversational recommender systems (RS) via retrieval-augmented generation (RAG). This flexible design introduces new sources of structural bias. This paper studies positional bias in top-1 and single-turn LLM-based RS: systematic dependence of the recommended item on the order in which candidates are presented in the prompt. We propose two complementary detection methodologies: (i) POB-ICM, based on contingency matrix analysis and Chi-square independence testing, and (ii) POB-UNBP, a sample-efficient approach derived from a uniformity null model under permutation-based evaluation, enabling stochastic decoding with formal statistical guarantees. Using a controlled RAG testbed for product recommendation, built on a real product catalog, a fixed prompt template and 2000 template-generated user queries with 20 permutations each (40,000 instances), both methodologies reject the null hypothesis of positional independence and uniformity in all evaluated state-of-the-art models. We introduce three complementary metrics -- permutation sensitivity, positional bias, and strong positional bias -- to quantify this effect, the latter targeting cases where the position displaces the recommendation towards clearly less relevant items rather than only choosing among comparably relevant ones. A robustness analysis across candidate-set size and decoding temperature shows that the bias is structural and model-dependent, not an artifact of deterministic decoding or candidate-set size. These results establish positional bias as a measurable, reproducible property of prompt-mediated recommendation pipelines in the evaluated setting and motivate routine auditing and governance controls (e.g., order randomization, standardized prompting) for deployed LLM-based RS. Source code is publicly released.