Sleep Tracking Apps and Devices — What They Actually Measure, and What They Guess
A wrist watch, a finger ring, a mat under your mattress, or an app listening through your phone's microphone — each shows you a neat sleep-stage graph the next morning. The problem is that a large part of that graph isn't a measurement at all, it's algorithmic guesswork, and a study comparing eleven popular trackers against polysomnography shows just how much that guesswork can vary.
No popular consumer device — watch, ring, under-mattress mat, or phone app — measures sleep directly. The gold standard in sleep diagnostics remains polysomnography (PSG): a simultaneous recording of brainwaves (EEG), eye movements (EOG), and muscle tone (EMG) under lab conditions. Consumer trackers have no direct access to any of these signals — instead, they rely on indirect proxies: movement (accelerometer), heart rate and its variability (a photoplethysmography, or PPG, sensor), and sometimes breathing or ambient sound.
From these indirect signals, an algorithm — usually proprietary and undisclosed — tries to reconstruct what PSG measures directly: when you fell asleep, how many times you woke up, and which sleep stage (light, deep, REM) you were in at any given moment. That reconstruction works reasonably well for some parameters and considerably worse for others — and that's exactly the boundary worth knowing before treating a morning graph as objective fact.
Accuracy of 11 Wearable, Nearable, and Airable Consumer Sleep Trackers: Prospective Multicenter Validation Study
Moderate evidence
Lee T, Cho Y, Cha KS, Jung J, Cho J, Kim H, Kim D, Hong J, Lee D, Keum M, Kushida CA, Yoon IY, Kim JW · JMIR mHealth and uHealth · 2023
A multicenter validation study compared 11 commercially available sleep trackers (5 wearables, including watches and the Oura ring, 3 "nearable" bedside devices, and 3 "airable" sound-based apps) against simultaneous polysomnography in 75 participants. Researchers collected a total of 3,890 hours of tracker recordings and 543 hours of polysomnography, analyzing 349,114 30-second sleep epochs. Sleep-stage classification accuracy (measured by macro F1 score) ranged from 0.26 to 0.69 depending on the device — meaning the gap between the weakest and best-performing hardware tested was more than twofold, and systematic biases (errors in one direction) differed by device category.
Three device categories — and why the distinction matters
Category
Examples
Main measurement signal
Wearable
sports watch, smartwatch, ring
wrist/finger movement + heart rate from a PPG sensor
Nearable (bedside)
under-mattress mat, nightstand sensor
body movement, sometimes breathing and sound, no skin contact
Airable (sound/app only)
phone app listening via microphone
breathing sounds, snoring, and movement in bed
Three types of sleep trackers from the validation study
This distinction isn't a purely technical curiosity — in the cited study, device categories differed not just in overall accuracy but in the direction of their errors: some systematically overestimated sleep efficiency, others underestimated nighttime wakefulness. That means two different devices worn the same night by the same person can show noticeably different numbers — and neither is necessarily "lying" more than the other; they're simply guessing differently based on different input signals.
What trackers measure fairly well
Parameters where consumer trackers usually do reasonably well
Overall total sleep time for the night — usually a reasonable approximation of PSG, especially for typical, uninterrupted sleep
Time of falling asleep and waking up — easy to detect from a sharp drop in movement and heart rate
Long-term trends in sleep efficiency for the same person, week to week — even if a single night carries error, the direction of change (better/worse than usual) tends to be useful
Resting heart rate and its overnight variability — PPG sensors are relatively reliable at rest, without movement
The common thread here is simple: these are parameters that can be inferred from movement and heart rate alone, without needing to distinguish subtle brain states. The closer the question is to "am I asleep at all," the better trackers perform. The closer it gets to "exactly which sleep stage am I in right now," the worse.
What trackers mostly guess, rather than measure
Sleep-stage breakdown (light/deep/REM) is the weakest point of most devices
Distinguishing light, deep (slow-wave), and REM sleep in polysomnography relies on brainwave patterns — a signal no consumer device has access to. Tracker algorithms estimate sleep stage from a combination of movement and heart rate variability, which in the cited study translated into clearly worse agreement with PSG than for total sleep time alone. The "deep sleep" percentage shown in an app is best treated as a rough algorithmic estimate, not a clinically measured value.
Parameters worth treating with the most skepticism
The exact percentage of REM and deep sleep on a given night — an algorithmic approximation, not an EEG measurement
Sleep latency (time to fall asleep) — hard to capture without a brain signal, especially when you're lying still but not yet asleep
The number and length of micro-awakenings during the night — easily confused with momentary stillness, and vice versa
The aggregate "sleep score" on a 0–100 scale — this is the output of many undisclosed assumptions baked into the manufacturer's algorithm, so comparing scores across brands is meaningless
Myth vs. fact
Myth
Since the app shows a detailed, minute-by-minute breakdown of REM, deep, and light sleep, it must have actually measured that.
Fact
That detailed graph is the output of a proprietary algorithm guessing sleep stage from movement and heart rate — not from brainwaves, as in polysomnography. The comparative study shows this classification accuracy is often moderate, sometimes clearly poor, depending on the device. A precise-looking graph is not proof of its precision.
This distinction has practical significance: it's easy to give a number on a phone screen the same status as a blood test result, when it's actually a fundamentally different kind of data — an indirect, algorithmic estimate, not a direct physiological measurement.
How to use this data wisely
Practical rules for interpreting sleep tracker data
Look at trends across many weeks, not a single night — one unusual night more often reflects measurement error than a real physiological change
Treat total sleep time and fall-asleep/wake-up times as the most reliable numbers in the whole dataset
Treat the sleep-stage breakdown and "sleep score" as an illustrative curiosity, not as a basis for health decisions
Don't directly compare numbers across different device brands — different algorithms, different definitions, different results for the same night
If tracker data raises a genuine health concern (e.g., suggesting sleep apnea, very low sleep efficiency for weeks), treat it as a prompt for a medical consultation and a possible referral for a real polysomnography study — not as a final diagnosis
When sleep tracking does more harm than good
Orthosomnia — an obsessive pursuit of a "perfect" sleep score
Sleep disorder literature describes a phenomenon called orthosomnia: a paradoxical worsening of sleep quality caused by excessive focus on tracker data and the drive for a perfect score. Someone lying in bed anxious that the app hasn't yet shown enough "deep sleep" makes it harder for themselves to fall asleep — a mechanism well documented in research on sleep-related anxiety, regardless of whether the app's data is even accurate.
If checking your sleep score right after waking up triggers frustration or anxiety more often than it provides useful information, it's worth checking the app less frequently or temporarily turning off score notifications — especially since, as the validation study shows, part of that score is only an approximation anyway.
Which type of device to choose
Type
Advantages
Limitations
Wearable (watch, ring)
continuous wear, plus daytime heart rate and activity
must remember to wear it at night and keep it charged; uncomfortable on the wrist for some people
Nearable (mat, bedside sensor)
zero body contact, nothing to wear or charge each night
less accurate when sleeping with a partner or pet in the bed
Airable (app + phone microphone)
no extra hardware, low entry cost
sensitive to ambient noise (partner, traffic, pets); one of the less accurate categories in the validation study
Pros and cons of each tracker type in everyday use
Limitations of this evidence
One study, specific devices, a specific population
Moderate evidence
The cited validation study covered 75 participants and 11 specific device models available at the time of the study — newer versions of the same hardware may have updated algorithms and different accuracy, so these exact numbers don't automatically transfer to a device bought today. Independent meta-analyses pooling multiple such studies, however, consistently point to a similar overall conclusion: consumer trackers are useful for a rough total sleep time, but clearly less reliable for sleep-stage breakdown — making this direction of the conclusion reasonably well supported, even if exact numbers vary between studies.
Our editorial recommendation
A sleep tracker is a useful tool for observing your own trends over time — not a substitute for a medical exam, and not a source of precise clinical data. It delivers the most value when treated like a diary rather than an oracle: it's worth watching whether sleep improves or worsens over weeks in response to lifestyle changes, rather than holding every single night accountable down to the minute of REM sleep.
The most common mistake I see is treating the morning "sleep score" like a blood test result — a precise, objective number. It's actually an algorithm's estimate, which even under controlled research conditions was sometimes clearly at odds with an actual sleep measurement.
Julia Wiśniewska, VitMode editorial team
Frequently asked questions
Yes, as long as expectations are realistic. For observing general trends — whether you're sleeping more or less than usual, how resting heart rate is changing — trackers are useful. For precise diagnosis of sleep disorders, such as sleep apnea or serious disruptions to sleep architecture, a medical exam is needed, ideally a doctor-ordered polysomnography.
The validation study didn't identify one category as universally best — accuracy varied between specific models within the same category, and each type had its own strengths and weaknesses depending on the parameter measured. Rather than looking for one "best" category, it's worth checking whether the manufacturer of a specific model has published independent validation studies.
Each manufacturer uses its own proprietary algorithm that interprets movement and heart rate signals differently. Differences in results between brands are normal and expected — they don't mean one device is "broken," just that both are estimating the same reality from different assumptions.
Some apps flag breathing irregularities or snoring, but they don't replace medical diagnosis of sleep apnea, which requires measuring blood oxygen saturation and airflow under controlled conditions. A signal from an app suggesting possible apnea is worth treating as a reason to see a doctor, not as a diagnosis.
Not necessarily. For some people, regularly and closely analyzing every night increases sleep-related anxiety, which can paradoxically worsen sleep quality — a phenomenon described in the literature as orthosomnia. If checking the app causes frustration more often than it gives useful information, it's worth reducing how often you look at it.
Not always, and not automatically — price and release date don't guarantee a more accurate sleep-stage classification algorithm. The only reliable way to assess this is checking whether a given model has been the subject of an independent validation study comparing it against polysomnography.
Total sleep time is usually clearly more reliable, since it relies on a simpler movement/stillness distinction. The percentage of deep and REM sleep requires a much more complex estimate from indirect signals, so it carries more error — as confirmed by the cited validation study.
MSc in Cognitive Neuroscience, host of a sleep-optimization podcast
Julia studied cognitive neuroscience planning an academic career, but partway through her PhD she realized she cared more about explaining research than running it. She started a podcast on sleep optimization — first for a handful of friends, now followed regularly by tens of thousands of listeners — and that podcast opened the door to writing for VitMode. She specializes in chronobiology, nootropics and recovery protocols, and her pieces often start from a question she asked herself during her own sleep experiments — including one memorable month living on a 28-hour "day," which she doesn't recommend anyone repeat. Off the clock, she sleeps surprisingly little for someone who writes about it professionally, and she's the first to laugh about it.
