Watches and rings that track sleep: what they get right and what they don't

What the device records and what it infers

The reference test for studying sleep, polysomnography, records the brain's electrical activity, eye movements and muscle activity. Using those signals, a technician classifies each 30-second segment of the night as wakefulness or as a specific sleep stage.¹⁵

A watch or ring records none of that. It mainly measures movement and heart rate⁷³ and, in some models, temperature as well;³ from there, an algorithm specific to each manufacturer infers whether you're asleep and what stage you're in.⁷ The app's “light,” “deep” and “REM” stages are therefore an estimate made from other signals, not a measurement of the same thing the reference test looks at.

Nor do all devices count the same things. Most lump the two lightest stages of polysomnography together as a single “light sleep,” and in a 2022 study, the app used to read the data from one of the watches called the sum of deep sleep and REM “deep.”⁴ That's why the numbers from two different devices can't be compared. Many models also show overnight data such as heart rate or heart rate variability, which have their own limits.

Asleep or awake: right about sleep, wrong about wakefulness

This is what they do best. In a lab study with 34 young, healthy adults, all the devices recognized as sleep 93% or more of the segments in which the person was asleep. They did much worse with the stretches when the person was awake: the four wrist-worn devices recognized between 18% and 54%.¹

The same pattern showed up at home, in 21 young adults who slept for a week with four wrist or finger devices, compared with a headband that records brain activity: they recognized between 35% and 45% of wakefulness.² And also in Korea, with four watches and a ring tested in a total of 75 adults with sleep complaints: between 23% and 48%.³

The reason is that, to a motion sensor, lying still in bed without sleeping looks a lot like being asleep.⁴³ The authors of one of the studies believe that the devices that detect wakefulness best also rely on heart rate.⁴

As a result, they tend to add sleep and subtract awakenings, and they are more wrong on bad nights: the worse the person sleeps, the further they stray from the reference test.¹³ In the lab study, two of the four wrist-worn devices added an average of about 45 minutes of sleep per night (44 and 47). The most accurate one was right on average, with a difference of 2.6 minutes, but on any given night the difference could range from 42 minutes too few to 47 too many.¹

At home there's another source of error: the device decides on its own when you went to bed and when you got up. In the home study it got this right on most nights, but not all, and the authors suspect that lying awake and relaxed in bed may make it start counting sleep too early.² A review commissioned by the Sleep Research Society also points to problems recording sleep outside the main overnight period⁶ (a nap, for example).

Sleep stages: the weakest estimate

When you move from “asleep or awake” to stages, the results get worse. In the lab study with six devices, the five worn on the wrist or finger agreed with polysomnography on the stage of each segment between 50% and 61% of the time. Once the agreement expected by chance was taken out, the result was low to moderate (a kappa of 0.20 to 0.44), below the 0.78 that, according to the authors, who cite another study, technicians scoring a polysomnography reach among themselves.⁴ In the Korean study, the four watches and the ring got between 54% and 65% of segments right.³

The errors aren't spread at random: what the device fails to recognize usually ends up counted as light sleep, whether it's wakefulness, deep sleep or REM.¹³

On top of that, a good average can hide very different nights. In the lab study, the three wrist-worn devices that gave stages estimated, on average, an amount of deep sleep similar to polysomnography's, but on any given night the difference could be around an hour, either way. The authors warn that, with so much variability, a similar average may be due to errors in one direction and the other canceling each other out.¹

In practice, and as our own deduction from these studies, a night with “little deep sleep” in the app says very little about how much deep sleep you really got.

Who these results apply to

A meta-analysis of 24 studies, with 798 people and watches and wristbands from several brands, found differences from polysomnography in total sleep time, sleep efficiency, time taken to fall asleep and nighttime awakenings, with results that varied widely from one study to another. It didn't analyze stages. Its authors conclude that these devices are useful for following the general pattern of sleep, as long as they are interpreted with caution.⁵

In that meta-analysis, most participants were healthy adults,⁵ and three of the four validation studies in this article were done with young, healthy adults.¹²⁴ The American Academy of Sleep Medicine points out that a validated device may not be validated for all ages, for people with sleep disorders or other illnesses, or for people taking certain medications.⁷ And the review commissioned by the Sleep Research Society notes that some personal characteristics, such as skin color or obesity, can make the sensors work less well.⁶

What it can be useful for

Within these limits, the most useful data are the simplest: what time you go to bed and get up, how long you spend in bed, and how that changes from one week to the next. The authors of the home study conclude that these devices are useful, for now, for tracking sleep and wakefulness, not stages.² Regular sleep times matter, as we explain in how much to sleep and how regularly, and a watch can help you see, for example, how far apart your weekday and weekend schedules are.

• Look at weeks, not nights. The error on any given night can be large, especially when you've slept badly.
• Always use the same device. Each one counts stages in its own way.
• Pay more attention to your schedule than to stages. That's what they record best.
• Check it against how you feel. Writing down each morning what time you think you fell asleep and how you feel gives you a second reference.

Many apps sum everything up in a “sleep score.” Each manufacturer defines it in its own way,⁷ so the same number doesn't mean the same thing on two devices, and it's calculated from the same data we discussed above. If a number catches your eye, before changing anything it helps to frame it as a question and a simple test.

Sometimes the problem isn't the number but chasing it. A series of three cases seen at an insomnia clinic described people who spent too much time in bed so that the device would show more hours; according to the authors, this may have made their insomnia worse, and the fact that these devices tend to overcount sleep can reinforce that habit.⁸ These are only three cases and they don't tell us how many people it happens to; where the name given to this pattern comes from and how to recognize it is covered in when measuring gets in the way.

What it can't tell you

A watch or ring doesn't diagnose sleep disorders. In 2025, the American Academy of Sleep Medicine warned that apps and devices that estimate the risk of sleep apnea, including some authorized for that purpose by the US Food and Drug Administration (FDA), are no use for diagnosing it or for ruling it out.⁹ For the FDA itself, tracking sleep trends is a wellness function; a product intended to screen for or diagnose a disease no longer is.¹⁰

A reassuring number doesn't rule anything out, either. If you sleep badly over a long period or something about your sleep worries you, the person to turn to is a health professional, not the number on your watch; in when habits aren't enough we explain when it's worth seeing one.