On the conversational persuasiveness of GPT-4
Bibliographic Data
| ID | 4698530 |
|---|---|
| Authors | Francesco Salvi (0009-0001-6884-6825, Fondazione Bruno Kessler, corresponding author), Maria Helena Ribeiro (0000-0002-6159-9657, Princeton University), Rosalia Gallotti (0000-0002-8088-1973, Fondazione Bruno Kessler), Robert West (0000-0001-7305-3654, Ecole polytechnique fédérale de Lausanne) |
| Year | 2025 |
| Volume | 9 |
| Issue | 8 |
| Pages | 1645-1653 |
| Publication date | 2025-05-19 |
| Peer Reviewed | Yes |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | Nature Human Behaviour (JOURNAL) |
| Journal identifiers | ISSN: 2397-3374 • E-ISSN: 2397-3374 |
| Publisher | Springer Science and Business Media LLC (PUBLISHER) |
| DOI | 10.1038/s41562-025-02194-6 |
| PMID | 40389594 |
| OpenAlex | W4410496075 |
| Language | EN |
| Citations received | 15 |
| References cited | 54 |
Early work has found that large language models (LLMs) can generate persuasive content. However, evidence on whether they can also personalize arguments to individual attributes remains limited, despite being crucial for assessing misuse. This preregistered study examines AI-driven persuasion in a controlled setting, where participants engaged in short multiround debates. Participants were randomly assigned to 1 of 12 conditions in a 2 × 2 × 3 design: (1) human or GPT-4 debate opponent; (2) opponent with or without access to sociodemographic participant data; (3) debate topic of low, medium or high opinion strength. In debate pairs where AI and humans were not equally persuasive, GPT-4 with personalization was more persuasive 64.4% of the time (81.2% relative increase in odds of higher post-debate agreement; 95% confidence interval [+26.0%, +160.7%], P < 0.01; N = 900). Our findings highlight the power of LLM-based persuasion and have implications for the governance and design of online platforms
Adversary · Computer security · Odds · Personalization · Persuasion · Persuasive communication · World Wide Web · Computer Science · Hate Speech and Cyberbullying Detection · Misinformation and Its Impacts · Psychology · Social Psychology · Topic Modeling
Playing with the dials of belief
Beyond efficiency
Plausibility, persuasion, and truth
Programmable support
The ethics of break-up chatbots
‘Yes, I Am a Slave to AI’
A Two-Step, Multidimensional Account of Deception in Language Models
Enhancing situation awareness for disaster policymaking via LLM-generated narratives
Personalized Persuasion Through Conversational AI
Large language models as a conduit for value shifts in contemporary China
Toward an artificial deliberation? On Google DeepMind’s Habermas Machine
Is Undisclosed LLM Use Morally Wrong
The Homily in the Algorithmic Age
Responsible AI and human collaboration in tourism management
The impact of advanced AI systems on democracy
Partial Proportional Odds Models for Ordinal Response Variables
Durably reducing conspiracy beliefs through dialogues with AI
Attitudes and Persuasion
Computer-based personality judgments are more accurate than those made by humans
Private traits and attributes are predictable from digital records of human behavior
Data quality in online human-subjects research
Taxonomy of Risks posed by Language Models
A Review and Conceptual Framework for Understanding Personalized Matching Effects in Persuasion
The small effects of political advertising are small regardless of context, message, sender, or receiver
A new readability yardstick.
One size does not fit all
Psychological targeting as an effective approach to digital mass persuasion
Human heuristics for AI-generated language are flawed
Personality, Gender, and Age in the Language of Social Media
Influencing Myself
Ordinal Regression Models in Psychology
Longitudinal data analysis using generalized linear models
Using a Personality-Profiling Algorithm to Investigate Political Microtargeting
The relationship of opinion-strength and order of self-produced arguments to number of arguments produced and opinion change
All the News That’s Fit to Fabricate
Large Language Models Can Argue in Convincing Ways About Politics, But Humans Dislike AI Authors
Analyzing ordinal data with metric models
Targeted Campaign Appeals and the Value of Ambiguity
Automatic personality assessment through social media language
Persuasion and Propaganda
| Unique citing works | 15 |
|---|---|
| Citations per year | 15 |
| Citation span | 2025 - 2026 (2) |
| Citation velocity | current |
| Highly cited | No |
| Citation types | Neutral: 13 |