Investigating perceived trust and utility of balanced news chatbots among individuals with varying conspiracy beliefs

Summary

This paper introduces and evaluates “Infobot,” a balanced news chatbot designed to expose users to an equal mix of mainstream and alternative/conspiratorial perspectives on climate change, with the goal of piercing information bubbles. Across two preregistered survey experiments, the authors adapt the Technology Acceptance Model (TAM) to chatbots and test whether users with high versus low conspiracy beliefs differ in their acceptance of such a tool. Contrary to the expectation — drawn from literatures on motivated reasoning and reactance — that conspiracy believers would reject a chatbot offering counter-attitudinal content, high-conspiracy users trusted, liked, and intended to use Infobot more than low-conspiracy users, suggesting mainstream-leaning users may themselves be a meaningful barrier to balanced information exposure.

Key Contributions

• Adapts the Technology Acceptance Model to balanced news chatbots, isolating perceived usefulness and trust as the dominant predictors of attitude and intention to use.
• Provides counterintuitive evidence that conspiracy-believing users can respond more favorably to balanced information tools than mainstream-believing users.
• Introduces a working depolarization prototype (“Infobot”) built on Rasa, presenting curated mainstream and alternative climate articles.
• Reframes the echo-chamber debate by suggesting selective engagement is bidirectional, not exclusive to conspiracy believers.
• Shares preregistration, materials, and SEM analyses via OSF for replication.

Methods

Two preregistered online experiments with US panel participants used Infobot to present 8 climate headlines (4 mainstream, 4 alternative) in a randomized carousel; participants read ≥4 article summaries. Study 1 (n=177) split groups via the Generic Conspiracist Beliefs Scale; Study 2 (n=58) used a custom 4-item climate-conspiracy scale (α=.83). TAM constructs (ease of use, usefulness, risks, trust, attitude, intention) were measured on 7-point Likert scales and modeled via SEM in lavaan with robust ML estimation, supplemented by Welch’s t-tests and exploratory analyses of article-selection and reading-time behavior.

Findings

• Perceived usefulness (β=.64 / .80) and trust (β=.35 / .40) predicted attitude, explaining ~91% of variance; ease of use and perceived risk were non-significant.
• Attitude predicted intention to use almost perfectly (β=.98 / .97; R²≈95%).
• High-conspiracy participants reported significantly higher trust (5.48 vs. 4.39 in Study 1; 5.54 vs. 4.54 in Study 2), more positive attitudes, and stronger usage intentions.
• High-conspiracy participants were more politically conservative, but in Study 1 did not differ on belief in anthropogenic climate change.
• Reading-time data suggested selective engagement in both directions: Study 1 participants overall read more mainstream articles, while Study 2 participants read more alternative ones; high-conspiracy users in Study 1 spent less time on most mainstream articles.
• A negatively worded trust item cross-loaded onto perceived risk and was dropped.

Connections

This paper sits alongside other recent attempts to use scalable interventions — chatbots, messaging, persuasion experiments — to shift politically charged beliefs. It connects most directly to Gauthier2026-iq and Szabo2026-rd on climate-message reception across ideological lines, and to Voelkel2026-lc on interventions targeting polarization and democratic attitudes. Its TAM-based focus on trust as a mediator distinguishes it from media-exposure work like Allcott2025-jb, but shares the broader question of whether engineered information environments can move entrenched audiences.

Podcast

A research-radio episode discusses this paper: Listen