Google’s Pixel Watch caught 2 in 3 cases of high blood pressure without a cuff

A Google smartwatch, reading nothing but the pulse light and motion sensor on the back of the wearer’s wrist, picked out about two in every three people with high blood pressure during a week of ordinary wear. It needed no arm cuff and no calibration, and in a head-to-head test against the gold-standard measurement it found more of those people than the single cuff check taken in a doctor’s office, according to a study by Google Research scientists and a physician at Beth Israel Deaconess Medical Center in Boston, published in the journal Nature.

That matters because high blood pressure rarely announces itself. More than a billion people live with it, and over half of the adults who have it do not know, which is why doctors call it a silent killer. A watch that nudges people toward a proper check could reach the ones who never see a clinician, although the team is clear that what it built is a screening prompt, not a diagnosis.

The system, which the authors call AI-PPG-ACC-HTN, does not try to read out a blood pressure number. It looks instead for the lasting signature of hypertension in the shape of thousands of pulse waves collected over several days on a Pixel Watch 3, and then gives each wearer a single probability that their pressure is too high.

Recommended ReadingPixel Watch 5’s AI Health Guardian Ships Before the FDA Clears It

In a trial of 196 adults at five sites in the United States, that probability flagged 65.8% of the people who turned out to have hypertension, compared with 55.3% for a standard office reading. Both methods were right about 90.0% of the people who did not have it. And when the watch raised a flag, it was correct four times out of five.

How a watch reads blood pressure without squeezing your arm

Every smartwatch with a heart-rate feature already shines green light into the skin and measures how much bounces back. That signal, called photoplethysmography, rises and falls as each heartbeat pushes blood through the vessels of the wrist. Its exact shape carries information about how stiff the arteries are and how the pulse wave reflects back from the rest of the body. The watch’s accelerometer adds a second signal: during moments of stillness it can pick up the tiny recoil of the body each time the heart ejects blood into the aorta, a trace known as a ballistocardiogram. The model also takes the wearer’s height, as a stand-in for how far the pulse has to travel.

The Google team first trained a neural network on 492,347 hours of unlabelled sensor data from 2,109 people who had taken part in earlier Fitbit studies. The network learned to turn every 15-second slice of pulse and motion data into a compact numerical fingerprint without ever being told anyone’s blood pressure. For each person, those fingerprints were averaged across every day of wear and fed to a simple classifier.

The averaging is the point. Optical pulse readings are easily thrown off by cold skin, a loose strap, the angle of the wrist or a sudden movement, which is why earlier cuffless devices had to be recalibrated against a cuff and could only track changes from that baseline. Pooling days of data lets those passing errors cancel out, in the same way that a 24-hour cuff monitor smooths over individual bad readings. Performance rose with the amount of data and began to level off after about an hour of usable signal per day.

To test it, the researchers enrolled 296 adults with no known heart disease who were not taking blood pressure drugs. Each wore the watch for a week, on average 14.7 hours a day, plus an ambulatory monitor that inflated a cuff around the opposite arm at regular intervals for 24 hours. They also took home cuff readings morning and evening, and had their pressure measured at the clinic. Hypertension was defined by the ambulatory monitor as an average of 130/80 mmHg or higher across the day and night, and 38.8% of the 196 people with complete data met that bar.

When the team looked at what the model was responding to, the pulse waves of people it rated high-risk were slightly narrower and rose more slowly, and the wave reflected back from the body was smaller relative to the first one. Those are known markers of stiffer arteries. Their ballistocardiogram traces also showed deeper dips at one point in the cycle, a pattern linked to resistance in the blood vessels.

The doctor’s cuff misses more people than you might expect

The comparison with the clinic is less flattering to the cuff than most patients would assume. A single office visit, judged against the 24-hour reference with the usual clinic threshold of 140/90, caught just over half of the people with hypertension. Doctors know why: some patients tense up in the exam room, a white-coat effect that inflates readings, while others have masked hypertension that only shows up at home or at night. A model that used only age, sex and body mass index caught 34.2%.

The researchers also tested the watch as a second opinion. Among 48 participants whose first office reading was high, it confirmed true hypertension in 78.4% of cases, and 96.7% of its positive calls were right. A repeat office visit did worse on both counts. At least three days of home cuff readings remained the strongest option, detecting 89.2% of cases.

The most obvious rival is Apple‘s hypertension notification feature, cleared by the US Food and Drug Administration, which watches optical pulse data over a month. Its supporting study reported 41.2% sensitivity and 92.3% specificity. That figure was measured against home cuff readings in a different group of people, so the two numbers are not a fair race, and the Google authors say only that their approach “may offer improvements” in catching people who would otherwise be missed.

Two details from the error analysis stand out. Of the 26 people with hypertension that the watch missed, 17 were also missed by the office cuff. And 9 of its 12 false alarms went to people whose pressure was raised by some other measure, often at night. The team found no link between the model’s mistakes and age, sex, body weight or skin tone, which matters because optical sensors have a history of working less well on darker skin. Just over a third of the analysed volunteers had dark skin tones on an instrument-measured scale.

What a two-in-three hit rate leaves out

The headline number cuts both ways. A watch that finds two in three people with hypertension also misses one in three, so a watch that stays silent is no reason to skip a check-up. The misses cluster at the milder end: the system caught 89.5% of people with a 24-hour average of at least 145/90 but only 52.8% at 125/75.

The advantage over the office cuff is also less certain than it looks. The confidence intervals overlap, from 54.0% to 76.3% for the watch and from 43.4% to 66.7% for the cuff, so the study shows the two are comparable rather than proving the watch is better.

The test group had clear limits too. People already taking blood pressure medicine or living with heart disease were excluded on purpose, so nothing here says whether the watch can track treatment. A third of enrolled volunteers dropped out of the analysis, 75 of the 100 because their ambulatory monitor data were missing or incomplete, often because they took the cuff off to sleep. Everyone was in the United States, and only one watch model was tested.

The study was funded by Alphabet, Google’s parent company, and 15 of its 16 authors are Alphabet employees. The authors themselves list the next hurdle: randomised trials showing that the alert finds people who would otherwise go undiagnosed, and that this improves their health. “Smartwatch-based hypertension screening is not intended to replace standard clinical arm cuffs or provide a definitive standalone medical diagnosis,” Paolo Di Achille and Ming-Zher Poh of Google Research wrote in a company blog post about the paper.

Common questions about smartwatch blood pressure checks

Can a smartwatch measure blood pressure without a cuff?

Not as a precise reading. Systems like this one use the watch’s optical pulse sensor and accelerometer to estimate whether someone’s pressure is persistently high over days, rather than giving a systolic and diastolic number. A positive result is meant to send the wearer to a validated cuff measurement.

How accurate is smartwatch hypertension detection?

In this study the Google system caught 65.8% of people with hypertension and correctly cleared 90.0% of those without it, measured against 24-hour ambulatory monitoring. About four in five of the people it flagged really had high blood pressure.

What is 24-hour ambulatory blood pressure monitoring?

It is a portable cuff worn for a full day and night that inflates automatically at regular intervals. Because it averages dozens of readings in everyday life, including during sleep, doctors treat it as the most reliable way to confirm hypertension.

What blood pressure counts as hypertension?

The threshold depends on how it is measured. The study used a 24-hour average of 130/80 mmHg or higher, a single clinic reading of 140/90 or higher, and a home average of 135/85 or higher, following international guidelines.

The paper appeared online in Nature on 5 November 2026 as an accelerated preview, after a preprint was posted in December 2025. Google says the work helped inform a blood pressure trends feature in its Google Health app, while describing the methods in the paper as exploratory research. The authors say validation in larger and more international groups, followed by randomised trials, comes next. Until then, the cuff stays on the arm, and the watch’s job is to get people to put it there.

Reference: Di Achille et al., “Opportunistically Detecting Signs of Hypertension on a Consumer Smartwatch,” Nature, 2026. DOI: 10.1038/s41586-026-11336-3

Tags: Google, medicine, blood pressure, hypertension, Pixel Watch, smartwatch

More Like This

Google’s $99 Fitbit Air has no screen — and that’s the point

Google’s Fitbit Edge gets a replaceable battery and loses tap-to-pay

Stay connected to your phone… and your blood sugar level

Pixel Tag: Google’s $29 AirTag Rival Won’t Ship Until November

Samsung Galaxy Glasses: What They Look Like, and Where the Footage Goes

Google CEO Sundar Pichai drops Google’s AI-code percentage and says self-improving AI is far off