How Explanation Framing Shapes Reliance on AI in Clinical Decision Support

In high-stakes clinical environments, even the most advanced AI systems can only augment—not replace—human expertise, but their effectiveness hinges on how their recommendations are communicated. This chapter dives into a groundbreaking study that explores how the framing of AI-generated explanations shapes diagnostic accuracy, professional confidence, and reliance behavior among obstetric nurses. Through a controlled experiment involving 52 participants and 15 clinical scenarios, the research compares gain-oriented, cost-oriented, and neutral framings to reveal surprising insights: while both valenced framings improve diagnostic accuracy over neutral explanations, cost-oriented language—such as emphasizing the risks of non-compliance—dramatically increases appropriate reliance on AI advice, often at the cost of heightened automation bias. The study uncovers how these framings influence not just final decisions but also the psychological biases that drive overreliance or skepticism, including automation bias (the tendency to defer to AI even when wrong) and self-anchoring bias (the tendency to dismiss correct AI advice due to overconfidence in initial judgments). With practical implications for AI designers, the findings challenge the assumption that neutral, fact-based explanations are always superior, demonstrating instead that strategically framed AI messages can enhance performance—provided the AI itself is reliable. The chapter also highlights the delicate balance between leveraging AI for better outcomes and safeguarding against unchecked automation bias, a critical consideration for safety-critical fields like obstetrics. For professionals seeking to optimize AI integration in clinical workflows, this research offers evidence-based strategies to craft explanations that not only inform but also responsibly guide decision-making.