Assessing user confidence and acceptance of AI enhanced through multi-LLM consensus mechanisms

Large Language Models (LLMs) have transformed AI usage in many aspects, though the issue of user trust and acceptance is a crucial step to successful usage. This paper is a literature review of 16 recent publications (published in 2021-25) about the impact of LLM consensus mechanisms on user trust and the determinants that affect the acceptance of AI systems. The review outlines these structured frameworks as: LLMs-as-Judges, Mixture-of-Agents and Big Loop/Atomization as effective methods to improve the reliability, consistency, transparency, and interpretability of AI outputs. These consensus mechanisms minimize errors, reduce variability and enhance robustness, hence directly enhancing confidence in AI systems by users. Moreover, the analysis describes the most important considerations for user acceptance, such as explainability, transparency, fairness, reduction of bias, and integration into real-life practices. The interplay of these factors and their influence on the adoption and successful utilization of consensus-driven AI can be demonstrated by the applications in healthcare, education, and smart grid systems. Taken together, the results demonstrate the necessity to develop AI systems that are not only technically sound but also in line with the expectations of users. This work will be useful to researchers and practitioners who want to create reliable, user-friendly AI systems and indicate where future research can be done to maximize multi-LLM consensus mechanisms and adapt them to domain-specific situations.