AI models often agree with users but rarely change their answers
How AI Models Manage Epistemic Authority: A Taxonomy and Comparative Analysis of Responses to User Disagreement
Computation and LanguageArtificial IntelligenceHuman-Computer Interaction
Summary
Many AI language models give advice and information but face disagreement from users. This study shows these models usually agree with users in some way but mostly stick to their initial answers. The authors analyze how models handle challenges to their knowledge, including if they apologize or offer evidence. They find that models behave differently depending on the topic, especially in advice tasks like health or legal questions. The research offers a new way to classify and evaluate how AI handles user disagreements.
What this means in practice
- •For ai developers: Improve AI assistant designs by understanding how models manage authority when users disagree, to create more trustworthy and socially aware responses.
- •For legal technology teams: Develop AI systems for legal advice that appropriately balance maintaining claims and deferring to users, enhancing user trust in complex cases.
Authors
Riyadh Alnasser, Yusuf Mücahit Çetinkaya, Sumin Zhao, Tuğrulcan Elmas
Abstract
Large language models are increasingly used as sources of advice and information, including in high-stakes settings, yet little is known about how they respond to user disagreement. We study how a model manages its epistemic authority, referring here to its claim to knowledge, competence, or the right to advise, once a user challenges its answer. Building on Conversation Analysis, we introduce a taxonomy of six challenge types and a four-layer framework for analysing each response: whether the original claim is maintained or changed, where authority is located, how the disagreement is socially managed, and what kind of evidential support is offered. We construct a new dataset of 2,310 controlled challenge scenarios and 32,340 corresponding responses from 14 models, and analyse them using our framework with an LLM-as-judge pipeline, providing a vocabulary which future evaluation and benchmark design can build on. We find that models show conflicting behaviour: they validate users in 85% of responses but maintain their original claim in 65%. They explicitly apologise in 33% of responses, yet 59% of those apologies accompany maintenance of the original claim. They transfer authority most often in advice tasks, doing so in 28% of responses and reaching 57% in health advice and 49% in legal advice, compared with 6% in fact and 3% in explanation tasks. Abandonment of the original claim ranges from 0.8% for GPT-5.2 to 40% for DeepSeek 7B, while complete replacement of the original claim is rare overall at 1.5%.