Trust in Generative AI for Health Information Consumption and the Effect of Learned Dependency: An Experimental Study

A randomized experimental study of 338 participants reveals that users who develop learned dependency on generative AI for health information exhibit weaker trust calibration and increased susceptibility to incorrect outputs. While information accuracy generally increases trust in AI-generated health content, highly dependent users show diminished ability to discern accuracy, and visual attention cues failed to mitigate this overtrust vulnerability.

This research addresses a critical gap in understanding how behavioral patterns interact with AI adoption in high-stakes domains like healthcare. The study demonstrates that repeated reliance on generative AI creates a psychological dependency that undermines users' native ability to evaluate information quality—a phenomenon with significant implications for public health outcomes and AI safety.

The findings emerge at a crucial moment when generative AI systems are increasingly deployed for medical information retrieval without adequate safeguards. Prior research highlighted accuracy concerns with large language models in healthcare contexts, but this work isolates how user behavior amplifies rather than moderates these risks. The positive correlation between learned dependency and trust suggests users develop heuristic shortcuts that bypass critical evaluation when they've grown accustomed to AI assistance.

For healthcare providers, insurers, and AI developers, these results indicate that naive interventions like visual attention cues prove insufficient. The significant interaction effect between accuracy and dependency reveals that dependency fundamentally alters how users process information—more dependency doesn't just increase baseline trust, it specifically weakens their ability to calibrate trust according to actual quality signals. This creates a compounding vulnerability where dependent users become progressively less capable of identifying false information.

Looking forward, the healthcare AI sector must prioritize dependency-aware design patterns rather than relying on presentation-layer solutions. Implementation of verification mechanisms, confidence scoring transparency, and user education about dependency risks should precede broader deployment. Regulatory bodies overseeing medical AI applications should consider these dependency effects when establishing guidelines for AI-assisted diagnosis and patient education.

• →Learned dependency on generative AI weakens users' ability to calibrate trust based on information accuracy.
• →Visual attention cues alone cannot mitigate overtrust in incorrect AI-generated health information.
• →Highly dependent users show significantly reduced trust discrimination between accurate and inaccurate outputs.
• →Repeated AI reliance creates psychological patterns that undermine critical evaluation of medical information.
• →Healthcare AI deployment requires dependency-aware design beyond simple UI interventions.