Emotional and Literacy Adaptation in Large Language Model-Based Medical Guidance: Development and Usability Study

Large language models (LLMs) are increasingly deployed as conversational agents for communication and guidance, but their ability to adapt responses to a patient's emotional state and literacy level remains insufficiently characterized. In patient-facing scenarios, effective communication requires not only accurate medical content but also appropriate tone, clarity, and responsiveness to user vulnerability.

This study aimed to evaluate, within a controlled, single-scenario experimental design, the ability of contemporary LLMs to adapt medical guidance responses to different levels of emotional sensitivity and literacy, and to assess their consistency in both generating and interpreting such adaptations.

A controlled benchmark was designed to evaluate how 4 LLMs adapt medical guidance responses to predefined emotional and literacy levels for a realistic patient profile asking an open-ended question. To establish a robust performance baseline, one model was deployed locally (Gemma 3 27B), and 3 are cloud-based (Claude Opus 4.5, Gemini 3 Pro, and GPT 5.2). Each model was tested across all combinations of 5 emotional levels, 3 literacy levels, and 3 prompt-verbosity variants, yielding 45 responses per model. Generated responses were assessed against human-reviewed ground truth for emotional and literacy alignment. A separate self-evaluation procedure measured each model's ability to interpret its own outputs using the same adaptation framework. The study also examined generation-interpretation consistency and quantitative text properties, with particular attention to response length and verbosity.

Literacy adaptation was consistently accurate across models, often achieving near-zero error, indicating strong capability in adjusting explanatory complexity. In contrast, emotional adaptation showed greater variability and was more sensitive to model architecture and prompt design. The larger cloud-based models demonstrated more stable performance overall, while the local model exhibited higher variability, particularly under shorter prompts. A systematic asymmetry was observed between generation and interpretation. The models frequently misclassified low emotional and literacy levels as "normal," suggesting limitations in recognizing user needs. Quantitative analysis revealed that response length increased with both emotional sensitivity and literacy level, with prompt verbosity and model-specific tendencies acting as secondary factors.

Within our scenario, current LLMs demonstrated the capacity to support adaptive medical guidance, but their performance was uneven across dimensions, especially in terms of emotional generation and interpretation. For health care applications, these systems should be treated as assistive tools under human oversight rather than as autonomous communicators, with careful consideration of ethical, privacy, and safety implications in sensitive health contexts.