Distilling arguments: A study of human and LLM persuasion in the online discourse

Persuasion is a key component of human interaction and has been extensively studied by computer scientists. Recent breakthroughs in AI agents, particularly large language models (LLMs), have opened up unprecedented avenues of online interaction, where users are increasingly exposed to AI-generated content. This makes it imperative to study the dynamics of persuasion in online communities and assess the effects LLMs might have on them. In this thesis, we address three crucial areas. First, we examine online user debates and characterize the persuasive strategies employed across different topics. Next, we evaluate the ability of state-of-the-art LLMs to generate and detect persuasive content. Finally, we address the potential misuse of LLMs towards influencing public opinion and explore summarization-based mitigation strategies. Our findings show that humans’ persuasive strategies vary significantly across topics, and LLMs demonstrate a reasonable understanding of persuasive content. Further, we find that LLMs tend to emphasize factual elements when summarizing arguments, although the results differ highly with the topic. We believe our work offers valuable insights into the dynamics of online persuasion and contributes to building robust guardrails against AI-generated persuasive content.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Gurjar, Omkar

Contributors dc:contributor

• Chandrasekharan, Eshwar

• Large Language Models
• Online Persuasion
• Argument Summarization

Statement dc:rights

• Copyright 2025 Omkar Gurjar

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/129251

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F129251