When Personalization Misleads: Understanding and Mitigating Hallucinations in Personalized LLMs

TL;DR

This paper investigates how personalization in large language models can lead to hallucinations by distorting factual responses, and proposes a method to mitigate this issue while preserving personalization.

Contribution

It introduces FPPS, a novel inference-time technique to reduce factual distortions in personalized LLMs, and presents PFQABench, a benchmark for evaluating factual and personalized QA.

Findings

FPPS significantly improves factual accuracy in personalized LLMs.

Personalization can cause models to generate answers aligned with user history rather than facts.

The proposed approach maintains personalized behavior while enhancing factual reliability.

Abstract

Personalized large language models (LLMs) adapt model behavior to individual users to enhance user satisfaction, yet personalization can inadvertently distort factual reasoning. We show that when personalized LLMs face factual queries, there exists a phenomenon where the model generates answers aligned with a user's prior history rather than the objective truth, resulting in personalization-induced hallucinations that degrade factual reliability and may propagate incorrect beliefs, due to representational entanglement between personalization and factual representations. To address this issue, we propose Factuality-Preserving Personalized Steering (FPPS), a lightweight inference-time approach that mitigates personalization-induced factual distortions while preserving personalized behavior. We further introduce PFQABench, the first benchmark designed to jointly evaluate factual and personalized question answering under personalization. Experiments across multiple LLM backbones and personalization methods show that FPPS substantially improves factual accuracy while maintaining personalized performance.