How does a smartwatch measure heart rate?
By shining light into the skin and measuring how much comes back — a technique called photoplethysmography, which measures blood volume changes rather than electrical activity, and which explains both what these devices do well and where they fail.
The mechanism. Green LEDs illuminate the wrist, and a photodiode measures reflected light. Blood absorbs green light, so as each heartbeat pushes blood through the capillaries, reflected light dips. Counting those dips gives heart rate. Green is used because it is strongly absorbed by blood and relatively unaffected by skin tone variation compared to some alternatives, though effects remain.
Why it fails when it does:
Motion. Movement changes the optical path and swamps the signal. Algorithms use accelerometer data to subtract motion, which works well for steady rhythmic activity and poorly for irregular movement — which is why weight training, rowing and racquet sports produce the worst readings, and why wrist devices frequently lock onto cadence rather than heart rate during running.
Fit. A loose watch admits ambient light and moves. Snug, one finger above the wrist bone, is the single biggest improvement available.
Cold, which reduces peripheral blood flow.
Tattoos, which absorb light and can prevent readings entirely.
Rapid changes, where averaging lags behind interval efforts.
What the other sensors do: red and infrared LEDs estimate blood oxygen saturation by comparing absorption at two wavelengths; electrodes on some watches take a single-lead ECG, which measures electrical activity and is a genuinely different and more clinically meaningful measurement; and skin temperature and electrodermal sensors feed stress and cycle estimates.
What is genuinely useful: resting heart rate trends; atrial fibrillation notifications, which have real evidence and have prompted diagnoses; and relative change over time.
What to treat cautiously: absolute accuracy during exercise, where a chest strap measuring electrical activity remains substantially more accurate; derived metrics like stress and recovery scores, which are proprietary and unvalidated; and blood oxygen readings, which are not medical-grade on most devices.