Pixel Watch will flag insulin resistance. It will not measure your blood sugar

The Pixel Watch 5 will not print a glucose reading. Google repeats that in footnotes for a reason. What it will start doing this month is file a monthly trend — high or not — on insulin resistance and blood pressure, built from heart rate, sleep and motion rather than a cuff or a blood draw. The same software is coming to Pixel Watch 3, Watch 4 and Fitbit Air. You do not need the new hardware. You do need a month of wear, and you should not treat the result as a lab test.

Google's launch post says Health Guardian works "quietly in the background 24/7" to track "usually invisible health and wellness indicators, such as insulin resistance trends." Florence Thng, the company's health intelligence product management director, carries the byline on it. The same copy says the insulin feature "analyzes multi-week physiological data to track shifts in your metabolic health without requiring a single drop of blood." The same pages call the features "intended for general wellness and informational purposes only," "not a prescreener for diabetes, hypertension or sleep apnea," and not tools "to adjust medications or treatment plans."

Brandon Ballinger, who in 2018 helped train an early neural net to spot diabetes signals in consumer heart-rate data, unpacked Google's numbers on Empirical Health. The company is using SensorFM, a foundation model pretrained on wearable data from five million opted-in users. The published validation used an earlier version of that approach, LSM-2, scored against HOMA-IR — an estimate from fasting insulin and fasting glucose, not the laboratory clamp that endocrinologists treat as gold standard.

Wearable data alone, tested on 72 people who were not in the training set, reached an AUROC of 0.75. Adding age, BMI, fasting glucose and lipids pushed that to 0.88. Ballinger's translation is the one buyers need: 0.75 means that if you hand the model one person with insulin resistance and one without, it ranks the right person higher about 75 percent of the time. It does not mean three-quarters of individual flags are correct. The independent group was small — 19 people with insulin resistance — and they wore a Fitbit Charge 6, with an average BMI of 30.6. HOMA-IR itself can swing 23.5 percent between two draws on the same person.

Google published a related accuracy discussion in Nature and says SensorFM was pretrained on more than a trillion minutes of data. That is a serious research stack. It is still a monthly label in the Google Health app, not a replacement for a Freestyle Libre or a fasting panel. One in five people in Google's study who had normal glucose already met the HOMA-IR threshold the model used. That is the interesting clinical idea: catching metabolic strain before glucose breaks. It is also why the footnotes exist.

The blood-pressure story is parallel. Google says the watch passively measures "your pulse and motion patterns throughout the month to estimate your blood pressure trends without requiring a cuff or manual calibration." Rishi Chandra, vice president for Google Health, told WIRED the company has years of opted-in Fitbit PPG and "an understanding of different conditions," then matches those signatures against arm-cuff data.

Daniel Jones, former president of the American Heart Association, told the same magazine there are no wrist devices the AHA recommends for monitoring blood pressure. Arm cuffs remain the gold standard for diagnosis and for guiding medication. The AHA did say late last year that cuffless wearables "show great promise" and have "the potential to transform" hypertension care. That is a research posture, not a green light to skip the cuff.

Samsung already offers on-wrist blood pressure with monthly cuff calibration. Apple shipped hypertension notifications on selected Watch models last year. Google's version is a trend, not a millimetre-of-mercury reading, and it is not for people who are pregnant. Barbara Eichorst, vice president of health care programs at the American Diabetes Association, told WIRED that consumer wearables are not a replacement for established blood-pressure measurements and that there are no ADA recommendations for using them to monitor insulin resistance. "The important consideration is not simply having access to more health data, but whether the information is clinically meaningful, understandable, and can appropriately support diabetes management," she wrote.

Google's product pages put both trend features in September 2026. Chase the Summit, after a couple of days with Watch 5 hardware in New York, told viewers the blood-pressure report is "not FDA cleared yet," arrives as a monthly trend rather than a daily number, and unlocks after about a month of wear. Insulin resistance, he said, is inferred from "bio signals, your heart rate, and your activity." Jordan Floyd at 9to5Google, two weeks into a Watch 5 review, listed the same Health Guardian summaries as still waiting on a software update, alongside strength-training guidance that was shown on stage and missing from the wrist.

Sandeep Waraich, Google's senior director of wearables, told WIRED the company spent years defending heart-rate accuracy for exactly this moment. "It might have felt strange—why are we harping about heart rate accuracy so much—back then," he said. "We kept investing in that, and now it's very improved, because all these large sensor models can detect these signatures, insulin resistance, blood pressure trends—but underneath, they need a really capable raw signal."

Google Health Coach, which is supposed to turn those flags into advice, needs a Google Health Premium subscription. The trend rows themselves are wellness features, not a diagnostic device. If a monthly "High" appears, Ballinger's advice is the boring one: talk to a doctor, who can order fasting glucose, insulin, HbA1c and a proper cuff. Illness, bad sleep or a new medication can shove wearable signals around. The Watch 5 got better at reading a pulse. It is not a lab on a strap, and Google's own footnotes are the most useful part of the launch.