How Accurate Is Wearable Sleep Tracking, Honestly?
Patricia Bedoya
·
·
6
min read
Cover image generated with AI — © Chubby Studio S.L.
KEY TAKEAWAY
Wrist sleep trackers detect sleep with roughly 95 percent sensitivity but detect wakefulness with only about 50 percent specificity, so quiet time in bed reads as sleep. Pooled across 24 validation studies, the typical gap against a sleep lab is around 17 minutes of total sleep time and about 5 percentage points of sleep efficiency. That is good enough to follow your own trend, and not good enough to judge a single night.
Every sleep tracker prints a number as if it were read off an instrument. It was not. It was inferred, from a wrist, by a model that never saw your brain. The honest question is not whether that inference is right, it is which parts of it are right and which are close to guesswork.
Here is the summary before the detail. Wrist trackers are excellent at noticing that you are asleep and mediocre at noticing that you are awake. That single asymmetry explains almost every complaint people have about their sleep data.
What does accurate even mean for a sleep tracker?
Validation studies put a device on someone who is also wired up for polysomnography, the sleep lab standard: brain waves, eye movement, muscle tone. Then they ask three separate questions. Did the device get the total right? Did it spot the minutes you were awake? Did it label the stages correctly? A device can do well on the first and badly on the other two, and most of them do.
Which is why the headline accuracy figure is close to useless. You are genuinely asleep for the large majority of the time a tracker is measuring, so a device that simply declared sleep from lights out to alarm would still score somewhere around 85 to 90 percent. Accuracy is inflated by the fact that the correct answer is usually sleep. The interesting number is the one nobody puts on the box.
How good is it at telling sleep from wake?
Very good in one direction. In a controlled study of 15 healthy adults, wrist actigraphy scored sleep correctly about 95 percent of the time. Wake was another matter: specificity was 54 percent on a normal night. Overall accuracy came out at 91 percent, which sounds excellent, and is carried almost entirely by the easy half of the problem.
Modern devices with a pulse sensor have not changed that shape much. A 2020 study validated minute-by-minute scoring on a consumer wearable against polysomnography in 41 participants and found sleep sensitivity of 95.5 percent, wake specificity of 55.6 percent, and overall accuracy of 90.3 percent. Total sleep time came out 18 minutes too high on average.
The part worth remembering is what happened in that first study when the nights got harder. As wakefulness increased, specificity fell to 45 percent, and to 37 percent after caffeine, dragging accuracy down to 72 percent. The measurement degrades exactly when the night is worth measuring. Lying still and awake at four in the morning looks, from the wrist, almost identical to sleeping.
Read the trend, not the night
FitWoody reads your sleep against your own history and shows it next to your metabolic baselines and training load, which is the comparison the data can actually support. Plus comes with a 7-day free trial.
Download the app
★★★★★
4.5 · 732 ratings on the App Store
How far off is the total, in minutes?
A 2025 meta-analysis pooled 24 validation studies covering 798 participants across Fitbit, Garmin, Apple Watch, WHOOP and others. The typical gap against the sleep lab was around 17 minutes on total sleep time, roughly 5 percentage points on sleep efficiency, about 13 minutes on time awake after falling asleep, and a couple of minutes on how long it took to drop off. The paper does not state its sign convention explicitly, so treat those as sizes of disagreement rather than proof of a direction.
Put that in context. A quarter of an hour of error is nothing across a month and everything across one night. If you slept seven hours and the device says six hours forty-five, your week is unaffected. If you are trying to work out whether last night was genuinely worse than the night before, that same error is bigger than the difference you are looking for.
It is also worth knowing what those studies look like. Validation runs in a lab, usually for a single night, on volunteers who agreed to sleep wired to an EEG: 15 healthy adults in one of the studies above, 41 participants with one overnight each in another. Young, healthy, monitored, sleeping in an unfamiliar bed. Those are the conditions under which the published figures were produced, which makes them roughly the device's best case rather than a description of your Tuesday.
Why a high correlation does not mean a good device
This is the trap in most marketing claims, and the sleep research community wrote a whole paper about it. A 2021 framework for testing sleep technology points out that correlation only describes linear association, not agreement: two measurements can correlate at 0.94 while one of them is systematically 50 minutes off. Their own worked example shows a correlation and an intraclass coefficient of 0.94 that would flatter a device which could not be used to calibrate anything.
What a serious validation reports instead is bias with limits of agreement, a Bland-Altman plot, and epoch-by-epoch analysis with sensitivity and specificity per stage. If a company quotes you a single accuracy percentage and nothing else, it has told you almost nothing about whether the device works.
So what is wearable sleep tracking actually good for?
It is good, genuinely, at the things that depend on the easy half of the problem. When you went to bed, when you got up, and how those two drift across weeks are the measurements you can lean on. Consistency of timing, total sleep as a rolling average, and whether this month looks like last month are all well within what a wrist device can support.
It is not good at telling you that you were awake for 43 minutes last Tuesday, at ranking one night against the night before, or at splitting a night into stage percentages you can act on. And it cannot diagnose anything. Sleep apnea, insomnia and the rest are clinical diagnoses made with clinical tools, and no wrist number substitutes for a conversation with a doctor.
How to read your own numbers with this in mind
Change the unit you look at. A single night is mostly noise; a two-week average is signal. If you want to know whether something you changed worked, give it a fortnight and compare averages, not mornings.
Judge the number against yourself rather than against a target. The measurement error is largely systematic, which means it mostly cancels out when you compare your own weeks with each other, and does not cancel at all when you compare your number with someone else's. That is the same reason your own baseline beats a generic normal range everywhere else in health data.
And treat stage percentages as the softest part of the report. If you want to know what those stages are doing, we covered what light, deep and REM sleep actually do; just do not build a decision on one night's split. When the data does say you slept badly, the useful response is adjusting the day rather than the diagnosis.
Where FitWoody sits in this
FitWoody reads what your Apple Watch already recorded, so it inherits these limits rather than escaping them. What it does with them is the point: it reads your sleep against your own history and alongside your metabolic baselines, where a rolling comparison is the honest use of the data. The right response to a noisy measurement is to stop reading it one night at a time, not to pretend the noise is not there.
Frequently asked questions
How accurate are sleep trackers compared to a sleep lab?
They detect sleep well and wake poorly. Validation studies typically report sleep sensitivity around 95 percent and wake specificity near 50 percent, with overall accuracy around 90 percent that is inflated because most of the night really is sleep. Pooled across 24 studies, the typical gap on total sleep time was about 17 minutes.
Why does my watch think I was asleep when I was lying awake?
Because quiet wakefulness and sleep look almost identical from the wrist: little movement, low and steady heart rate. This is the single biggest weakness of wrist sleep tracking, and it gets worse on broken nights, when specificity in one controlled study fell from 54 percent to 37 percent.
Can a sleep tracker diagnose a sleep disorder?
No. Sleep apnea, insomnia and other sleep disorders are clinical diagnoses made with clinical tools. A tracker can reasonably prompt you to raise something with a doctor, but the number itself is not a finding and should not be treated as one.