Google researchers have developed an investigational system that turns wearable signals into a “coachable health age” and simulates how changes in sleep, steps and exercise could affect it over 12 weeks. The score uses modifiable measures such as resting heart rate, daily steps, sleep duration and active minutes rather than trying to predict a person’s birth year.
A sensor world model was trained on 2.28 million person-weeks from roughly 100,000 consenting Fitbit users. Given two weeks of baseline data, it forecasts 20 physiological markers under different routines. An optimizer weighs projected improvement against a person’s historical flexibility and safety limits, then a language model turns the selected path into a coaching plan.
In a blinded review of 49 profiles, five lifestyle coaches preferred the simulation-grounded plan in 71.4% of comparisons. Five physicians rated 99.2% of those plans within safe exercise-intensity limits and 97.6% accurate on physiological claims. A separate analysis linked an older-than-calendar health score with higher later risk of hypertension, cardiovascular disease or type 2 diabetes.
The work remains retrospective. Google explicitly describes the models as research prototypes, not clinical diagnostic tools. Blood-test extensions were evaluated on 819 participants, while coaching evaluations used offline profiles. Prospective trials must still show whether the advice changes behavior and health outcomes in everyday use.