Patients searching for a reliable home blood pressure monitor often turn to popular AI chatbots for advice. They expect quick, accurate guidance on which devices meet clinical standards. New findings released today show that trust is misplaced.

Three of the four leading AI tools correctly identified whether a monitor had passed independent validation testing only 63% to 83% of the time. Google Gemini performed somewhat better, yet still erred in 9% to 14% of queries. The results come from preliminary research presented at the American Heart Association’s Hypertension Scientific Sessions 2026 in Arlington, Virginia.

American Heart Association researchers tested 324 popular home blood pressure monitors available in Canada. Of those, 145 had received validation through recognized protocols. The remaining 179 had not. Devices were drawn from three independent registries — StrideBP, ValidateBP and Hypertension Canada — plus lists of known non-validated units and top-selling models on Amazon in Canada, Australia and the United States.

The study, conducted in April and May 2026, posed standardized questions to ChatGPT, Microsoft’s Copilot, Perplexity AI and Google Gemini. Questions varied slightly in phrasing to test consistency. Answers were scored against the official registry data. Performance varied sharply by tool. Gemini answered correctly 86% to 91% of the time depending on exact wording. The other three models lagged noticeably.

“We found that most AI tools performed only slightly better than if you had flipped a coin for each question,” said Anna Soriano, M.D., a third-year internal medicine resident at the University of Montreal and the study’s presenting author. “Even Google Gemini, which performed best, was often wrong and couldn’t find information that is easily located.”

All four tools occasionally labeled validated monitors as unvalidated. They also showed inconsistency. When researchers repeated queries about devices that had produced mixed results, the same AI often gave different answers on different days or from different computers. Such variability undermines confidence for anyone seeking dependable product advice.

The implications stretch beyond consumer convenience. High blood pressure affects more than 125 million U.S. adults, roughly 47% of the population, according to the American Heart Association’s 2026 Heart Disease and Stroke Statistics Update. Only about one in four keeps readings below the target of 120/80 mm Hg. Inaccurate home monitors can produce faulty readings. Those errors may lead to missed diagnoses, unnecessary medication changes or false reassurance that blood pressure remains controlled.

The 2025 American Heart Association Guideline for the Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults explicitly recommends validated devices for home use. Patients can consult their clinician or visit independent sites such as validatebp.org. The new study reinforces that simple directive. Rather than ask an AI chatbot, check the registries directly.

Soriano and her colleagues noted that the validation information sits in free, public databases. In theory, advanced language models should retrieve and interpret it without difficulty. Their repeated failure to do so raises broader questions about current AI limitations in health-related product recommendations. Earlier coverage in Digital Trends had already flagged similar concerns based on prior analysis of chatbot performance.

Keith C. Ferdinand, M.D., FAHA, volunteer expert for the American Heart Association and vice chair of its 2025 high blood pressure guideline, reviewed the findings. He pointed to AI’s promise in areas such as cardiac imaging while stressing caution in direct clinical decision support. “This study highlights the necessity for caution when the technology is applied to clinical decision-making,” Ferdinand said, as reported in coverage by HyperAI.

Other recent research adds context to the mixed picture of AI in hypertension care. A 2025 study on large language models analyzing common hypertension scenarios found GPT-4 achieved 83% accuracy and 86% safety, still trailing human experts who scored 92% and 93% respectively. That work appeared in the journal Hypertension. Separate investigations into AI voice agents showed they can improve reporting accuracy among older adults and reduce clinician workload. Yet those successes involve structured interactions, not open-ended product advice.

Chatbots have improved in some medical domains. One analysis found ChatGPT outperformed Bing when answering questions about home hypertension management techniques. Another systematic review of chatbot interventions for hypertension patients reported gains in medication adherence and self-monitoring. Blood pressure reductions, however, remained inconsistent across trials.

Even so, the fresh AHA poster delivers a clear warning for everyday users. Blood pressure monitors appear on retail shelves in dozens of models. Many carry impressive-looking specifications. Without validation against established protocols, their readings cannot be trusted for clinical decisions. AI tools that cannot reliably distinguish validated from unvalidated devices risk steering consumers toward poor choices.

Researchers advise both clinicians and the public to bypass chatbots for this specific task. Confirm validation status through the independent registries. The process takes minutes. The potential health consequences of getting it wrong justify the extra effort.

Meanwhile, AI developers face a pointed challenge. Improving retrieval from authoritative public sources represents a solvable technical problem. Until models demonstrate consistent accuracy on straightforward verification questions, their role in guiding medical device selection should remain limited. Patients deserve better than coin-flip odds when selecting tools that directly influence cardiovascular care.

The study arrives as wearable and cuffless blood pressure technologies gain traction. Apple Watch hypertension notifications, cleared by regulators in 2025, have drawn scrutiny for possible false reassurance in lower-risk groups. Cuffless devices still lack full clinical validation standards for diagnosis and management, according to recent expert statements. In that environment, the importance of trusted upper-arm monitors grows rather than diminishes.

Soriano’s team plans further analysis of the data. They intend to examine why certain models fail to cite registry listings even when search results surface them. Future work may test updated model versions released after the April-May testing window. For now, the message remains direct. When it comes to picking a blood pressure monitor, skip the chatbot. Check the validated list instead.