AI Search Tools Misidentify Validated Blood Pressure Monitors in Preliminary Study

Asking an AI search tool whether a home blood pressure monitor has passed accuracy testing can produce the wrong answer—even when the device appears on an official registry. A preliminary study of ChatGPT, Microsoft Copilot, Perplexity and Google Gemini found that all four sometimes rejected monitors that had already met clinical validation standards.

The American Heart Association published the findings October 8, ahead of the study’s scheduled presentation that evening at its Hypertension Scientific Sessions 2026 in Arlington, Virginia. The research is a conference abstract, not a peer-reviewed paper, and its findings remain preliminary.

Three ways to ask the same question

Researchers tested 324 monitors in Canada during April and May 2026: 145 validated devices and 179 unvalidated ones. The sample drew from StrideBP, ValidateBP and Hypertension Canada registries, a list of known unvalidated devices, and top-selling monitors on Amazon in Canada, Australia and the United States.

Each tool received three question formats for every device. The researchers varied how much direction the prompt supplied, rather than testing only one casual question:

• A general, public-style question about whether the monitor was validated.
• A more specific question directing the tool to a named validation registry.
• A question with additional details that required a one-word answer.

Google Gemini performed best. Accuracy varied with question phrasing, and the reported ranges below should not be read as a single score shared by every product.

Reported answer accuracy

Google Gemini%

86%91%

Gemini’s correct answers ranged from 86% to 91%, depending on question wording.

ChatGPT, Copilot and Perplexity%

63%83%

Results for the other three tools ranged from 63% to 83% across tools and question formats.

Finding the registry was not enough

All four tools were less accurate at recognizing validated monitors than unvalidated ones. That difficulty extended to lesser-known devices, even though those monitors appeared on the same official registries as more popular products.

Presenting author Anna Soriano, an internal medicine resident at the University of Montreal, said the tools sometimes found an official listing but appeared unable to interpret it as proof of validation. The researchers could not determine exactly why the tools struggled with devices that had passed testing.

Consistency was another problem. Researchers retested 20% of devices that had generated mixed results, using three examiners on different computers and days. The same tool often gave a different answer to the same question. The April–May testing also makes these results a historical snapshot, not a measurement of today’s tools.

Check the device, not just the answer

Validation means independent, third-party testing has shown that a monitor gives consistently accurate blood pressure readings. The American Heart Association’s 2025 guideline recommends using a validated home monitor. Researchers advised patients and healthcare professionals to confirm a device’s status through recognized, free public registries.

Soriano warned that an incorrect AI answer could lead someone to believe an unvalidated device had passed testing. Inaccurate readings from such a device could then contribute to inappropriate diagnosis or treatment decisions. That was a potential consequence she identified, not a patient outcome measured in this study.