On the conversational persuasiveness of GPT-4
Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti, and Robert West · 2025 · corrected 2026 · Nature Human Behaviour
Original research ↗ (new tab)FRONTIERS / Comparative and American politics
A supplier offers a chatbot for a public transit consultation, backed by a certificate showing that stored demographic profiles are disconnected. As an adviser to Bridgehaven’s Civic Technology Commission, you must decide what evidence a license needs. Read two recent AI-persuasion studies, check six sentences against official notices, reconstruct a proposed causal process and defend an enforceable rule within a limited audit budget. Then apply your reasoning to a fresh public-service case. Keep experimental comparisons, factual reliability, citizen access and political judgment distinct while making them inform one decision. Bridgehaven and its local records are fictional.
Choose For fun / For class inside the activity. The same evidence and political problem remain available; class mode asks for the required responses. Save a local backup and a readable report before submitting through your course system.

Built-in explanations introduce the studies. For a seminar, use the reading companion before discussion. Familiarity with basic concepts helps; the readings make the research question explicit.
Self-contained inquiry for introductory students · No advance reading required · Work in class or across sessions; student completion time has not yet been measured.
Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti, and Robert West · 2025 · corrected 2026 · Nature Human Behaviour
Original research ↗ (new tab)Kobi Hackenburg, Ben M. Tappin, Luke Hewitt, Ed Saunders, Sid Black, Hause Lin, Catherine Fist, Helen Margetts, David G. Rand, and Christopher Summerfield · 2025 · Science
Original research ↗ (new tab)The reading companion introduces the studies and distinguishes their findings from the fictional case. Original articles retain their own copyright and access conditions. A programmed result is not a measured effect in the world.
This Possible States edition runs without student accounts or hosted class submissions. The study, decision, evidence and response tasks are retained. Your work stays on your device; a backup restores progress and a readable report can be submitted separately. Free educational use & licensing information →.
Created by Yoav Duman.