Fitbit's ai coach now eavesdrops on your blood sugar and sleep cycles
Your wrist just became a medical resident. Google’s Gemini-powered Health Coach inside Fitbit’s Public Preview can now read continuous-glucose-monitor graphs, cross-check them against last night’s sleep stages, and tell you—without a white coat—why that 2 a.m. ice-cream pint flattened your deep-sleep curve.
Sleep scores finally grow teeth
The headline tweak is a 15 % jump in sleep-stage accuracy, enough to separate lying-motionless doom-scrolling from actual REM. Public-preview testers already see the difference: same hours-in-bed, new truth. The algorithm listens to heart-rate variance and SpO₂ micro-oscillations the way a somnologist listens to breathing; the watch now labels wake epochs down to thirty-second slices. Translation: your “eight hours” could be six and change, and the coach will name the culprit.
Glucose is the second vein opened. Pair a Dexcom or OneTouch monitor and the chat window turns into a dietitian that never blinks. Ask “why the spike at 10:17?” and the bot surfaces the bagel you logged, overlays it on the glycemic curve, and suggests tomorrow’s swap. It’s not medical advice—Google stamps that disclaimer hard—but it is the first time consumer wearables fuse live glycemia with longitudinal sleep data without a prescription portal in between.
Medical records slip through the side door
US users can pipe lab PDFs via CLEAR’s identity layer; LDL, HbA1c, and TSH values become conversational variables. The coach won’t diagnose, yet it will flag that your resting heart rate crept up eight beats the same week your ferritin dropped, nudging you toward an appointment before fatigue becomes a cliff.
Women’s-health tracking remains the glaring ellipsis—cycles, ovulation, pregnancy metrics are still rerouted to the vanilla Fitbit app. Google admits the gap; users toggle between preview and production builds like bilinguals switching dialects.
Still, the shift is semantic. Fitbit began as a pedometer that guilted you into 10 000 steps; it is mutating into a silent resident who remembers every skipped workout, every rogue frappuccino, every restless night. The question is no longer whether the numbers are accurate—15 % sharper sleep staging sees to that—but whether we want a machine that never forgets how we fail. For now, the preview is opt-in; the data, supposedly, can be wiped. Yet the memory of your glucose spike at 3 a.m. lingers in Google’s cloud long after the caramel aftertaste is gone. The coach learns; deleting the log won’t erase the lesson.