Increasing Large Language Model Accuracy for Care-Seeking Advice Using Prompts Reflecting Human Reasoning Strategies in the Real World: Validation Study.

Background: Current prompting techniques for large language models (LLMs), such as ChatGPT, mainly focus on well-structured, low-uncertainty problems; yet, many real-world tasks (eg, care-seeking decisions) are ill-defined and involve high uncertainty. Naturalistic decision-making (NDM) specifically analyzes how humans make accurate decisions in such settings, but NDM concepts have not yet been applied to LLM prompt engineering. Objective: This study aimed to determine whether prompting strategies inspired by NDM (specifically based on recognition-primed decision-making and the data-frame theory) could improve LLM performance in a real-world, high-uncertainty task, such as making care-seeking decisions. Methods: We evaluated 10 ChatGPT models (GPT-4o, GPT-4.1, GPT-4.1 mini, o3, o4 mini, o4 mini high, GPT-5.1 Instant, GPT-5.1 Thinking, GPT-5.2 Instant, and GPT-5.2 Thinking) using 3 prompting strategies: a default prompt solely asking the LLMs to classify the case vignettes, a recognition-primed prompt tasking the models to reason according to recognition-primed decision-making, and a data-frame prompt tasking the models to apply the data-frame theory. The task was taken from a standardized and validated evaluation framework and instructed the LLMs to advise on the appropriate care-seeking action for 45 real patient case vignettes across 3 urgency levels (emergency, nonemergency, and self-care). Each model-vignette-prompt combination was tested 10 times to assess and account for output variability. Accuracy was analyzed using mixed effects logistic regression. Additionally, we evaluated accuracy for each urgency level and examined output variability. Results: Both NDM-inspired prompts increased overall model accuracy (recognition-primed: 67.6%; data-frame: 66.7%) compared to the default prompt (63.3%). The greatest improvements were observed for self-care recommendations, where accuracy increased from 13.4% (default prompt) to 29.8% (recognition-primed prompt) and 24.6% (data-frame prompt). Performance on 2 emergency and 30 nonemergency cases remained high across all prompts. Notably, NDM-inspired prompts made nonreasoning models start giving self-care advice, even though they rarely or never provided self-care advice with the default prompt. Output variability was similar across the 3 prompts. Conclusions: Using LLMs with prompts inspired by NDM, which are designed to reflect real-world human reasoning, improves the accuracy of LLMs in care-seeking tasks, particularly for self-care advice, without reducing performance in the included emergency or nonemergency cases. These findings indicate that NDM-inspired prompts can offer an advantage when LLMs are used for real-world decisions involving ambiguity and uncertainty. The impact of output that reflects real-world human reasoning on users' decision-making must be evaluated in future studies.