Must Read: A Systematic Survey of Computational Persuasion

TL;DR

This survey articulates AI-driven persuasion through three lenses—AI as Persuader, Persuadee, and Persuasion Judge—grounding the analysis in social-science theory while mapping computational progress. It introduces a three-part taxonomy—Evaluating Persuasion, Generating Persuasion, and Safeguarding Persuasion—and synthesizes approaches, datasets, and evaluation frameworks for both human and automated assessment, including long-context and multimodal considerations. The paper reviews practical applications in negotiation, debate, and jailbreaking, and emphasizes risks such as manipulation and unsafe content, proposing strategies like selective acceptance, red-teaming, and adversarial frameworks to mitigate harm. It concludes with forward-looking directions: standardized, multi-dimensional evaluation; scalable human-like assessment; adaptive, culturally aware persuasion; and rigorous safeguards to ensure safe, ethical deployment of persuasive AI systems.

Abstract

Persuasion is a fundamental aspect of communication, influencing decision-making across diverse contexts, from everyday conversations to high-stakes scenarios such as politics, marketing, and law. The rise of conversational AI systems has significantly expanded the scope of persuasion, introducing both opportunities and risks. AI-driven persuasion can be leveraged for beneficial applications, but also poses threats through manipulation and unethical influence. Moreover, AI systems are not only persuaders, but also susceptible to persuasion, making them vulnerable to adversarial attacks and bias reinforcement. Despite rapid advancements in AI-generated persuasive content, our understanding of what makes persuasion effective remains limited due to its inherently subjective and context-dependent nature. In this survey, we provide a comprehensive overview of computational persuasion, structured around three key perspectives: (1) AI as a Persuader, which explores AI-generated persuasive content and its applications; (2) AI as a Persuadee, which examines AI's susceptibility to influence and manipulation; and (3) AI as a Persuasion Judge, which analyzes AI's role in evaluating persuasive strategies, detecting manipulation, and ensuring ethical persuasion. We introduce a taxonomy for computational persuasion research and discuss key challenges, including evaluating persuasiveness, mitigating manipulative persuasion, and developing responsible AI-driven persuasive systems. Our survey outlines future research directions to enhance the safety, fairness, and effectiveness of AI-powered persuasion while addressing the risks posed by increasingly capable language models.

Figures (3)

• Figure 1: The three key perspectives of AI-based persuasion. (1) AI as Persuader: AI generates persuasive content to influence humans or other AI agents, which can be used for both beneficial and harmful purposes. (2) AI as Persuadee: AI systems can be influenced or manipulated, either by humans or other AI, leading to unintended, unethical, or harmful outcomes. (3) AI as Persuasion Judge: AI is used to assess persuasive attempts, identifying persuasive strategies, detecting manipulation, and evaluating ethical considerations.
• Figure 2: Taxonomy of computational persuasion.
• Figure 3: This survey categorizes the evaluation of persuasiveness into three main types: (1) evaluation of argument persuasiveness, (2) human evaluation of LLM-generated content, and (3) automatic evaluation of LLM persuasiveness. For argument persuasiveness, models are typically trained on human-annotated or naturally labeled data to assess the persuasive strength of given arguments. For evaluating LLM persuasiveness, two branches of research emerge: one uses human judges to rate AI-generated content or interactions, while the other relies on LLM-based or non-LLM automatic metrics to perform the evaluation.