How Accurate Is Gemini (a Large Language Model) in Providing Lifestyle Recommendations for Hypertension? A Comparative Analysis with Current Guidelines

Abstract

Objective: This study aims to compare the lifestyle modification (LM) recommendations provided by the generative artificial intelligence tool Gemini for a patient with home blood pressure measurements of 135/85 mmHg on two separate occasions, with current international (ESH/ESC) and national (TKD/THD) clinical hypertension guidelines.

Methods: The analysis focused on a home blood pressure value of 135/85 mmHg, corresponding to the threshold of Stage 1 Hypertension. Six major LM recommendations generated by Gemini—Salt Restriction, Diet, Physical Activity, Weight Control, Harmful Habits, and Stress Management—were identified and compared with the guidelines in terms of recommendation class (Class I/II), level of evidence (A/B), and estimated clinical effect size.

Results: All six lifestyle recommendations provided by Gemini were consistent with guideline-endorsed, evidence-based interventions (Class I, Level A/B) proven to effectively lower blood pressure. This finding indicates that the AI tool generally delivers accurate and reliable information on this critical clinical topic. However, several gaps were noted, including the absence of specific guidance on dietary models such as the DASH diet, the role of resistance training, and recommended follow-up intervals of 3–6 months for reassessment.

Conclusion: Generative AI has considerable potential to support health literacy and patient engagement in chronic disease management such as hypertension. Nevertheless, the comparison with established guidelines reaffirms that AI-generated lifestyle advice must always be personalized, detailed, and validated by qualified healthcare professionals to ensure clinical effectiveness and patient safety. Furthermore, the inherent risk of hallucination in large language models remains a significant limitation that warrants cautious evaluation in health-related contexts.