Grounded Persuasive Language Generation for Marketing

We introduce an AI-powered framework called "AI Realtor" designed to generate persuasive and factually accurate real estate marketing content. The framework uses large language models (LLMs) within a three-module agent: a Grounding Module that identifies marketable features, a Personalization Module that tailors content to user preferences, and a Marketing Module that ensures factual accuracy and local relevance. The paper details the construction of a realistic evaluation environment using real real estate data and human subjects, focusing on grounded persuasion, which is based on facts and audience tailoring. Experimental results show that AI Realtor's generated descriptions are significantly preferred over human-written ones, suggesting the potential for LLMs in large-scale, targeted marketing while emphasizing responsible generation. The authors also discuss evaluating AI-simulated feedback and checking for hallucinations to ensure trustworthiness.