This paper introduces a novel auction mechanism for aggregating preferences over Large Language Model (LLM) outputs, which provably converges to the theoretically optimal distribution. The mechanism is designed to balance the usefulness of LLM-generated replies to users with the preferences of self-interested advertisers. It ensures that truthful reporting is a dominant strategy for advertisers and aligns each advertiser's utility with their contribution to social welfare, making it a general-purpose solution for truthfully aggregating preferences over LLM-generated replies. The mechanism operates without LLM fine-tuning or access to model weights, and it can incorporate contextual information about advertisers, significantly accelerating convergence. Experiments with a publicly available LLM show that the mechanism significantly boosts advertiser value and platform revenue with low computational overhead. The mechanism is strategyproof, meaning that it is a dominant strategy for each agent to truthfully report her preferences, and it is equitable in the sense that each agent's utility is proportional to her contribution to social welfare. The mechanism is particularly well-suited for online advertising, where advertisers aim to influence the responses of an LLM to align with their interests, while platforms seek to maximize advertiser value and ensure user satisfaction. The mechanism can be applied in any setting with monetary transfers, making it a general-purpose solution for truthfully aggregating the preferences of self-interested agents over LLM-generated replies. The mechanism is also ex-ante individually rational, meaning that an agent is in expectation better off by participating. The mechanism is shown to be effective in generating significant value for participants while also effectively recapturing a considerable portion of this value as revenue.This paper introduces a novel auction mechanism for aggregating preferences over Large Language Model (LLM) outputs, which provably converges to the theoretically optimal distribution. The mechanism is designed to balance the usefulness of LLM-generated replies to users with the preferences of self-interested advertisers. It ensures that truthful reporting is a dominant strategy for advertisers and aligns each advertiser's utility with their contribution to social welfare, making it a general-purpose solution for truthfully aggregating preferences over LLM-generated replies. The mechanism operates without LLM fine-tuning or access to model weights, and it can incorporate contextual information about advertisers, significantly accelerating convergence. Experiments with a publicly available LLM show that the mechanism significantly boosts advertiser value and platform revenue with low computational overhead. The mechanism is strategyproof, meaning that it is a dominant strategy for each agent to truthfully report her preferences, and it is equitable in the sense that each agent's utility is proportional to her contribution to social welfare. The mechanism is particularly well-suited for online advertising, where advertisers aim to influence the responses of an LLM to align with their interests, while platforms seek to maximize advertiser value and ensure user satisfaction. The mechanism can be applied in any setting with monetary transfers, making it a general-purpose solution for truthfully aggregating the preferences of self-interested agents over LLM-generated replies. The mechanism is also ex-ante individually rational, meaning that an agent is in expectation better off by participating. The mechanism is shown to be effective in generating significant value for participants while also effectively recapturing a considerable portion of this value as revenue.