Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI Realtor: Towards Grounded Persuasive Language Generation for Automated Copywriting

Jibang Wu, Chenghao Yang, Yi Wu, Simon Mahns, Chaoqi Wang, Hao Zhu, Fei Fang, Haifeng Xu

TL;DR

The paper tackles grounded persuasion in automated copywriting by integrating an economic information-design framework with an LLM-based agent, AI Realtor, to produce fact-grounded, personalized real estate marketing text. It formalizes signaling via an attribute-feature mapping and models a three-module pipeline—Grounding, Personalization, and Marketing—that extracts marketable features, personalizes content to buyer preferences, and injects localized surprisal through Retrieval Augmented Generation. Empirical results from a Zillow-derived REM dataset and a real-world-style buyer interface show AI Realtor achieving a substantial win rate over human experts while maintaining high factual fidelity, supported by Elo-based evaluations and hallucination checks. The work demonstrates the viability of scalable, targeted copywriting that preserves factual grounding and offers a blueprint for extending signaling-informed design to other high-stakes domains, with reproducibility through open prompts and code.

Abstract

This paper develops an agentic framework that employs large language models (LLMs) for grounded persuasive language generation in automated copywriting, with real estate marketing as a focal application. Our method is designed to align the generated content with user preferences while highlighting useful factual attributes. This agent consists of three key modules: (1) Grounding Module, mimicking expert human behavior to predict marketable features; (2) Personalization Module, aligning content with user preferences; (3) Marketing Module, ensuring factual accuracy and the inclusion of localized features. We conduct systematic human-subject experiments in the domain of real estate marketing, with a focus group of potential house buyers. The results demonstrate that marketing descriptions generated by our approach are preferred over those written by human experts by a clear margin while maintaining the same level of factual accuracy. Our findings suggest a promising agentic approach to automate large-scale targeted copywriting while ensuring factuality of content generation.

AI Realtor: Towards Grounded Persuasive Language Generation for Automated Copywriting

TL;DR

The paper tackles grounded persuasion in automated copywriting by integrating an economic information-design framework with an LLM-based agent, AI Realtor, to produce fact-grounded, personalized real estate marketing text. It formalizes signaling via an attribute-feature mapping and models a three-module pipeline—Grounding, Personalization, and Marketing—that extracts marketable features, personalizes content to buyer preferences, and injects localized surprisal through Retrieval Augmented Generation. Empirical results from a Zillow-derived REM dataset and a real-world-style buyer interface show AI Realtor achieving a substantial win rate over human experts while maintaining high factual fidelity, supported by Elo-based evaluations and hallucination checks. The work demonstrates the viability of scalable, targeted copywriting that preserves factual grounding and offers a blueprint for extending signaling-informed design to other high-stakes domains, with reproducibility through open prompts and code.

Abstract

This paper develops an agentic framework that employs large language models (LLMs) for grounded persuasive language generation in automated copywriting, with real estate marketing as a focal application. Our method is designed to align the generated content with user preferences while highlighting useful factual attributes. This agent consists of three key modules: (1) Grounding Module, mimicking expert human behavior to predict marketable features; (2) Personalization Module, aligning content with user preferences; (3) Marketing Module, ensuring factual accuracy and the inclusion of localized features. We conduct systematic human-subject experiments in the domain of real estate marketing, with a focus group of potential house buyers. The results demonstrate that marketing descriptions generated by our approach are preferred over those written by human experts by a clear margin while maintaining the same level of factual accuracy. Our findings suggest a promising agentic approach to automate large-scale targeted copywriting while ensuring factuality of content generation.

Paper Structure

This paper contains 48 sections, 9 equations, 12 figures, 1 table.

Table of Contents

  1. Introduction
  2. A Benchmark for Grounded Persuasion
  3. Motivations and Challenges
  4. An Economic Scaffolding of Copywriting
  5. The Agentic Implementation of AI Realtor
  6. Grounding Module: Predicting Credible Features for Marketing
  7. Personalization Module: Aligning with Preferences
  8. Marketing Module: Capturing Surprisal via RAG
  9. Evaluations
  10. Evaluation by Human Feedback
  11. Evaluation through AI Feedback
  12. Hallucination Checks
  13. Related Work
  14. Discussion
  15. Data.
  16. ...and 33 more sections

Figures (12)

  • Figure 1: Illustration of the Design Pipeline of AI Realtor.
  • Figure 2: Illustration of the inductive feature schema construction pipeline.
  • Figure 3: Comparison of model performance using Elo ratings and win rates. Elo ratings represent overall persuasiveness, and win rates reflect relative persuasiveness. Both metrics are based on evaluations by human subjects.
  • Figure 4: Analyses of Simulating Human Feedback with AI Feedback.
  • Figure 5: Faithfulness Scores for Hallucination Checks.
  • ...and 7 more figures