Assimilation and Contrast in LLM--User Interactions

Read full text Summarise with VIRAT
Authors: Sebastián Vallejo Vera, Ernesto Calvo, Paula Clerici
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This pre-analysis plan describes the design, measurement strategy, and pre-registered hypotheses for an analysis of assimilation and contrast effects in Large Language Model (LLM)--user interactions. Drawing on the formal framework of Calvo and Chang (2025), we test whether respondents perceive the ideology of an LLM in ways that parallel how voters perceive the ideology of political parties: pulling ideologically congruent interlocutors closer (assimilation) and pushing ideologically incongruent ones further away (contrast). We embed these tests within an online survey experiment administered to a national sample of approximately 3,000 Canadian respondents, in which each participant completes one or two conversations with a randomized LLM on assigned policy topics and then places both themselves and the LLM on a standard 1--7 left--right ideological scale. We additionally pre-register an extension that tests whether prior exposure to a partisan social media post heightens both assimilation and contrast in subsequent LLM ideology perception.

Fields of Study & Subjects

Electoral Systems and Political Participation Ethics and Social Impacts of AI Opinion Dynamics and Social Influence

Publication details

Publication date
1 January 2026
Language
EN
Document type
Preprint