How accurate is sleep tracking?

Sleep labs use polysomnography, which records brain waves and more through sensors stuck to your head and body. A wrist device has movement and a pulse to go on. Compared with that, wearables do one thing well and several things poorly.

I don't test devices and I haven't worn any product named here. Every result below names its authors, year and sample, and it was checked as of 3 November 2026.

A tracker isn't a medical device. Wearable readings are estimates for general fitness and wellness, and they don't diagnose or treat any condition. If you're worried about your sleep, or you have symptoms or concerns, talk to a clinician. A sleep score isn't a sleep study.

Good at sleep, poor at wake

In study after study, trackers catch nearly all of your sleep. They're much worse at noticing you're lying awake. Researchers call those two sensitivity and specificity.

Lie still at 3 am, reading nothing and worrying about Tuesday, and your watch may well log it as sleep.

Chinoy (peer-reviewed, Sleep) is the cleanest on independence: the authors state no funding from or involvement by any device company, and the Office of Naval Research paid for it. It tested devices including the Fitbit Alta HR, Garmin Fenix 5S and Vivosmart 3. Devices did worse on nights of disrupted sleep, so the people most curious about their sleep may get the least reliable readings.

Schyvens 2025 (peer-reviewed, Sleep Advances; no financial arrangements stated) tested the Fitbit Charge 5, Fitbit Sense, Withings Scanwatch, Garmin Vivosmart 4, Whoop 4.0 and Apple Watch Series 8.

Sleep stages are estimates

Those deep, light and REM charts look exact. They aren't. Agreement with a lab is fair to moderate at best.

Miller and colleagues (2022, peer-reviewed, Sensors) put six devices on 53 adults aged 25.4 ± 5.9 for one night. The study was funded by the Australian Institute of Sport, and the authors' research group takes funding and equipment from WHOOP. WHOOP wasn't involved in this study, but you should know about the link.

Kappa adjusts for lucky guesses, and 1.0 would be perfect. Half right, in other words, for the first three. These are older generations, so don't read the table as a ranking of what's on sale now.

Others land in the same area:

• Schyvens 2025: kappa of 0.21 to 0.53 across six devices, with the Apple Watch Series 8 highest.
• Lee 2023 (peer-reviewed, JMIR mHealth): a macro F1 score (1.0 is perfect) of 0.26 to 0.69 across 11 trackers, in 75 patients at Korean sleep centres. Some authors hold stock in Asleep, which makes one tracker in the test. Those were clinic patients, so results may not carry over to healthy athletes.
• Robbins 2024 (peer-reviewed, manufacturer-funded, Sensors): 35 adults aged 20 to 50 with no sleep disorder, one night. The Oura Gen3 did not differ from the lab on stages. The Fitbit Sense 2 read light sleep 18 minutes over and deep 15 under, and the Apple Watch Series 8 read deep 43 under and light 45 over. Those are one-night averages, and single nights will land further off. Oura Ring Inc. funded the study, and Robbins sits on Oura's Medical Advisory Board and reports Oura consulting fees, so this isn't independent evidence for Oura.

I can't name a "most accurate sleep tracker" from this set. Nobody fairly can.

Pooled data says the same

Lee and colleagues (2025, peer-reviewed, J Clin Sleep Med) pooled 24 studies and 798 patients, and found wrist trackers differed significantly from a sleep lab on total sleep, efficiency, latency and wake after sleep onset. The mix includes old brands such as Jawbone and Basis. I've left the figures out because I couldn't confirm their units from the text I read.

Recovery scores: no gold standard

Recovery and readiness scores are proprietary summaries built from inputs like HRV and sleep. Nothing exists to validate them against. Lepley and colleagues (2026, peer-reviewed narrative review, Sensors) put it this way: "there is no gold standard", and "the composite scores themselves are largely unvalidated."

So I treat a score as a prompt to think about how you feel, and not as a measurement. I found no good evidence that one predicts performance or injury. Brands also build them differently, so a 70 on one band and a 70 on another aren't the same thing.

The gaps

Current firmware is the main one. Most papers don't report it, and makers change algorithms without notice. The newest models above, such as the Apple Watch Series 8 and Fitbit Sense 2, are several years old. I found no independent lab test of this year's bands against polysomnography, so I'm giving no figure for them.

What I'd use it for

Total sleep across weeks seems a fair thing to watch, on the same device, since sleep and wake detection is where trackers do best. Bedtime consistency works too, because you know whether you went to bed. I'd ignore a single night's deep-sleep figure, and I wouldn't let one bad score change a training session.

If you're choosing a device for this, see best wearables for sleep and recovery, or compare the bands in Whoop vs Garmin Cirqa vs Polar Loop. The sleep and recovery picks are a shorter route.