A new study has warned against relying on artificial intelligence-powered chatbots to select home blood pressure monitors, urging consumers to verify devices against independent validation registries before purchasing or using them.
اضافة اعلان
Asking an AI chatbot whether a home blood pressure monitor is reliable may produce an incorrect answer, according to new research raising concerns about the accuracy of these tools when providing medical information.
The study, presented at the American Heart Association’s Hypertension Scientific Sessions 2026, found that popular AI-powered search tools identified validated blood pressure monitors with accuracy rates ranging from 63% to 91%, depending on the tool used.
The findings suggest that relying solely on AI to verify the reliability of home medical devices could put users at risk of making inaccurate health decisions.
AI struggles to assess blood pressure monitors
In this context, “validation” refers to independent testing that confirms a blood pressure monitor’s ability to provide consistently accurate readings.
A device is considered validated when independent organizations, unaffiliated with the manufacturer, establish through testing that it meets the required accuracy standards. Three official registries track devices that pass these tests.
The study examined 324 blood pressure monitors, including both validated and non-validated devices.
Researchers tested four widely used AI tools: Google Gemini, Microsoft Copilot, ChatGPT and Perplexity.
They posed questions in three different formats to each tool to assess its ability to determine whether the monitors had passed independent validation tests.
Gemini performed better than the other tools, with accuracy rates ranging from 86% to 91%. The other three tools recorded accuracy rates between 63% and 83%.
However, all the tools struggled more to confirm that devices were validated than to identify non-validated monitors.
This occurred even when the devices were clearly listed in official registries accessible to the AI tools.
Inconsistent answers raise concerns
The researchers found that the problems extended beyond answer accuracy to include inconsistent results when questions were repeated.
When they asked the same AI tool the same question on a different day or using another computer, they sometimes received completely different answers about the same device.
Dr. Anna Soriano, one of the study’s authors, said most tools performed only slightly better than random guessing between two equally likely options. She added that Gemini, despite its comparatively stronger performance, repeatedly failed to find information that was readily accessible.
These inconsistencies are particularly concerning in healthcare, as incorrectly classifying an inaccurate monitor as validated could lead users to rely on faulty blood pressure readings.
Such readings could subsequently contribute to an incorrect assessment of a person’s health or inappropriate treatment decisions.
The findings highlight the need for caution when using AI tools for medical decisions, particularly when the information directly affects disease diagnosis and treatment.
Why do these findings pose a risk to users?
The accuracy of home blood pressure monitors is essential for health monitoring, as incorrect readings can influence decisions made by both doctors and patients.
Researchers warned that users’ trust in AI-generated answers could lead them to purchase devices that do not meet the required clinical standards, believing they had undergone tests confirming their accuracy.
The problem is that some AI tools may provide answers that appear credible without necessarily relying on accurate information about a device’s validation status.
For this reason, the researchers advised against relying on chatbots to verify blood pressure monitor validation at present.
Instead, they recommend consulting recognized registries directly, including the validated blood pressure monitor registry available at www.validatebp.org.
The study emphasizes that quick access to information through AI is not enough to guarantee its accuracy, particularly in medical matters that require reliable data and independent scientific testing.
Resource: Al-Ghad.
