JMIR Publications

ABSTRACT

Background:

The rise of generative AI tools such as ChatGPT is rapidly transforming how people access information online. In the health context, generative AI is seen as a potentially disruptive information source due to its low entry barriers, conversational style, and ability to tailor content to users’ needs. However, little is known about whether and how individuals actually use generative AI for health purposes, and which groups may benefit—or be left behind—in this emerging landscape, raising important questions of digital health equity.

Objective:

This study aimed to assess the current relevance of generative AI as a health information source and to identify key factors predicting individuals’ intention to use it. We applied the Unified Theory of Acceptance and Use of Technology (UTAUT 2), focusing on six core predictors: performance expectancy, effort expectancy, facilitating conditions, social influence, habit, and hedonic motivation. In addition, we extended the model by including health literacy and health status—two established predictors of online health information seeking. A cross-national design enabled comparison across four European countries.

Methods:

A representative online survey was conducted with N = 1,990 participants aged 16 to 74 years from Austria (n = 502), Denmark (n = 507), France (n = 498), and Serbia (n = 483). Structural equation modeling with metric measurement invariance was used to test associations across countries.

Results:

Usage of generative AI for health information was still limited: only 39.5% of respondents reported having used it at least rarely. Generative AI ranked last among all measured health information sources; instead, medical experts and online search engines are still the most frequently used health information sources. Despite this, performance expectancy (b range = .44–.53; all p < .001), habit (b range = .28–.32, all p < .001), and hedonic motivation (b range = .22–.45, all p < .001) consistently predicted behavioral intention in all countries. Facilitating conditions also showed small but significant effects (b range = .12–.24, all p < .01). In contrast, effort expectancy, social influence, health literacy, and health status were unrelated to intention in all countries, with one marginal exception (France: health status, b = −.09, p = .007). Model fit was good (CFI = .95, RMSEA = .03), and metric invariance was confirmed.

Conclusions:

Generative AI use for health information is currently driven by early adopters—those who find it useful, easy to integrate, enjoyable, and have the necessary skills and infrastructure to do so. Cross-national consistency suggests a shared adoption pattern across Europe. To promote equitable adoption, communication efforts should focus on usefulness, convenience, and enjoyment, while ensuring digital access and safeguards for vulnerable users.