How Do Wearables Actually Track Sleep Stages?

This piece covers how consumer wearables estimate sleep stages using motion and heart-rate signals, how that compares with polysomnography (PSG) and with dedicated EEG-based wearables, and what the validation research does and doesn't establish. It does not rank specific products or offer guidance on which device to buy.

Research comparing consumer wearables against polysomnography confirms that most wrist and finger devices estimate sleep stages from movement and heart-rate patterns rather than from brain activity itself. That indirect method produces measurable gaps against the PSG standard, gaps that narrow with newer algorithms but do not close entirely. Wearables built around actual EEG sensors, a distinct and much smaller category, show consistently high accuracy in comparison, which suggests the sensor type itself is a bigger factor than any single brand's software.

What Your Wearable Is Actually Measuring at 3 A.M.

Every sleep app hands over a tidy chart by morning: neat blocks of light, deep, and REM sleep stacked across the night, as if the device had been watching your brain the entire time. It hasn't. Most of the sensors people strap to a wrist or slide onto a finger are reading pulse and motion, then inferring which stage of sleep those signals most likely belong to. That gap between the confident-looking graph and what the sensor can actually detect is where a lot of the back-and-forth over tracker accuracy comes from.

I keep seeing the same handful of questions come up when people compare notes on their trackers, though the research doesn't always answer them the way you'd expect.

The Sensor Inside the Device Decides How Close the Guess Gets

Line those three findings up and a pattern shows itself. Fitbit models that rely mainly on movement plus heart rate, without a dedicated staging algorithm, tend to overcount how much sleep actually happened. Oura's newer algorithm, evaluated directly against multi-night polysomnography, closes most of that gap for total sleep time and even for light and deep sleep duration, but it still misses on sleep efficiency by a small but consistent margin. Meanwhile, wearables built with real EEG sensors, the same kind of signal a sleep lab reads, land in a different accuracy bracket altogether.

That last point matters more than it sounds. The stage labels themselves, N1 through N3 and REM, come from a scoring system built for actual brainwave recordings, the kind described in what REM sleep actually looks like on an EEG. A wrist-worn accelerometer and pulse sensor is trying to reverse-engineer those same labels from a much thinner slice of information. Some algorithms clearly do that better than others, but none of the accelerometer-and-heart-rate devices reviewed here claim to be reading the brain directly. Because they aren't.

Where the Studies Draw a Line

It's tempting to read 'Oura underestimated sleep efficiency by about 1 percent' and assume that number generalizes to everyone wearing the ring. The study behind that figure was conducted in generally healthy adults between 20 and 70 years old in a single country. It doesn't tell us how the same algorithm performs in people with diagnosed sleep disorders, in much older adults, in children, or across different populations entirely. The Fitbit meta-analysis has its own boundary too: it's built from studies of specific past-generation models, not necessarily the version currently on someone's wrist.

None of this means the numbers are wrong. It means they describe a specific device, a specific algorithm version, and a specific group of participants, and stage-tracking accuracy reported for one combination of those three things doesn't automatically carry over to another.

Why the Stage-Level Detail Might Matter Beyond the Score

Part of why people push so hard on stage accuracy is that deep sleep and REM aren't just trivia for a morning readout. Long-term wearable data from a large research program found that REM sleep and deep sleep were inversely associated with the odds of incident atrial fibrillation. And irregular sleep patterns tracked with higher odds of several chronic conditions over years of monitoring. That's a different kind of evidence than a device-versus-PSG validation study, it's about what stage-level data predicts over time rather than how precisely any single night is measured, but it helps explain why the accuracy question gets asked so often. If stage proportions are meaningfully linked to health outcomes, then how well a device estimates them starts to matter more than it would for a simple wake-versus-sleep count. Anyone curious about how deep sleep specifically fits into that picture might find the research on deep sleep amounts a useful next stop, as would anyone weighing sleep tracking against longer-term brain health questions covered in the research on sleep and dementia risk.

Common questions