06 · Research Seminar · 2025

Trusting AI with your health?

A 2×2 online experiment testing whether urgency and language style influence trust in fictional AI-generated health advice.

OwnershipSix-person team research project

This page presents the shared study. The supplied project files do not include a task-level contribution log, so no individual role is claimed.

Research question

Do the urgency of a health situation and the use of lay or medical language change how much people trust AI-generated health advice?

Approach

From question to evidence.

01
Designed four short fictional AI-health videos crossing urgent versus non-urgent situations with lay versus medical-jargon language.
02
Ran a pre-test with 54 people, then collected 121 usable main-study responses through Qualtrics.
03
Adapted ten items from the Wake Forest Physician Trust Scale; the reported internal consistency was α = .806. The team analyzed the 2×2 design in Jamovi.

What the work found

The study found no statistically significant main effect of urgency (p = .069), no main effect of language style (p = .059), and no interaction (p = .751). Trust was highest in the lay/non-urgent condition and lowest in the jargon/urgent condition, but those descriptive differences did not support the hypotheses.

Limitations & next step

The scenarios were hypothetical and online, the sample leaned young and highly educated, and a physician-trust scale had to be adapted for AI. The findings therefore support caution: they are useful for refining future studies, not for claiming that language and urgency never affect trust.

Evidence from the source work

Selected artifacts.

High-resolution poster panel explaining the urgency, language style, and trust variables in the AI health-advice experiment — Source visual: high-resolution methods panel showing how the two independent variables and trust outcome were operationalized.

High-resolution poster panel reporting the interaction test and key findings of the AI health-advice experiment — Source visual: the interaction result and concise key findings from the team poster.

High-resolution line chart comparing trustworthiness across urgent and non-urgent scenarios using lay and medical-technical language — Source visual: mean perceived trustworthiness across all four conditions. The descriptive pattern did not reach statistical significance.

Methods & tools used here

2×2 experimentQualtricsScale adaptationPre-testJamoviFactorial ANOVA