How accurate are wearables?

Sleep trackers are good at telling sleep from wake and at total sleep time. They are weaker at spotting time awake and at sleep stages.

• Evidence: Some
• Low effort
• Free
• Works: today

Why it matters

The reference test for sleep is polysomnography (PSG): a night in a lab with sensors on your head that record brain waves, eye movements and muscle tone. Sleep stages are defined by those brain waves. A ring or watch can’t see brain waves. It estimates sleep from movement, heart rate, heart rate variability and, on some devices, skin temperature.

So a tracker is always guessing at stages from indirect signals. How good the guess is decides how much weight you should put on each number in your app.

How to do it

1. Trust total sleep and timing most. These are what wearables measure best [2][4].
2. Treat time awake as a likely underestimate. Devices are weakest at spotting wake during the night [1][5].
3. Read stages as a trend, over a week or more. Single-night stage numbers are the least reliable [2].
4. Be more sceptical after a bad night. Devices did worse on nights with disrupted sleep [1].
5. If your tracker and your body disagree, believe your body. Feeling rested after a “poor” score is real information.

Common mistakes

• Assuming your exact model was tested. Most studies used older models: Miller tested the Oura Gen 2 and WHOOP 3.0, and Chinoy tested the Garmin Fenix 5S and Vivosmart 3 [1][2]. Software updates change the algorithms, so results for older models may not apply to yours, in either direction.
• Reading an average result as your result. A device can be right on average across a group and still be off by a lot for one person on one night.
• Ignoring who did the study. Some validation work is funded by, or done by people working with, the company that makes the device. That doesn’t make it wrong, but independent replication matters.

Effect on your sleep score

• Total sleep and sleep vs wake: in Chinoy’s lab study of 34 young adults, every device caught at least 93% of sleep (sensitivity ≥0.93) [1]. In Miller’s study of 53 adults, sleep-vs-wake agreement with PSG was 86–89% for Apple Watch, Garmin, Oura and WHOOP [2]. In Robbins’s study of 35 adults, sensitivity for sleep was 95% or more for Oura Gen3, Apple Watch Series 8 and a Fitbit [4].
• Wake and efficiency: specificity, the share of true wake the device caught, was only 0.18–0.54 in Chinoy’s study [1]. Kappa, a measure of agreement beyond chance (0 = chance, 1 = perfect), was 0.30–0.51 for sleep vs wake in Miller’s study [2]. An expert review lists misclassifying wake as a main limit [5].
• Deep sleep and REM: Miller found stage-by-stage agreement of 53% for Apple Watch, 50% for Garmin, 61% for Oura and 60% for WHOOP [2]. In Robbins’s study, Apple Watch underestimated deep sleep by 43 minutes on average, while Oura did not differ from PSG on any stage [4].
• Time to fall asleep: an Oura meta-analysis found no significant average difference from lab measures [6]. That is an average across people, not a per-night guarantee.

Evidence

Rated Some. Several lab studies against PSG, and a 2024 expert review for the Sleep Research Society, agree on the pattern: consumer devices now do better than older research wristbands (actigraphy) at sleep vs wake, but still misclassify wake and are inconsistent on stages [1][5].

Who funded what:

• Chinoy 2021 was funded by the US Office of Naval Research, with no company funding [1].
• Miller 2022 was funded by the Australian Institute of Sport. The authors’ lab receives research support from WHOOP, which was not involved in the study [2].
• Altini and Kinnunen 2021 were an Oura adviser and an Oura employee. Their model reached 96% accuracy for sleep vs wake and 79% for four stages in 106 people over 440 nights [3]. This was a research model tested on Oura data, not necessarily the algorithm in the app.
• Robbins 2024 was funded by Oura, and the first author sits on Oura’s medical advisory board [4].
• Khan 2025, a meta-analysis of 6 studies with 388 people, found no significant average difference between Oura and lab measures for total sleep, efficiency, wake, deep or REM sleep [6]. Average agreement doesn’t mean each night is accurate.

Limits: studies are small (34 to 106 people), mostly one to three nights in a lab, and mostly in healthy, often young, adults. A review notes devices may perform differently with darker skin or obesity [5].

General information for healthy adults, not medical advice. If you have a sleep problem that lasts, read when to see a doctor. Sources checked 4 October 2026. Spot a mistake? Tell us.