Sleep Trackers: How Accurate Are They Really?

A patient once showed me a year of sleep data, nightly graphs, stage percentages, a running score, and asked me what was wrong with her deep sleep. She slept well, felt fine, and had spent months worrying about a number her watch invented. Sleep trackers are remarkable engineering and really useful tools, but only if you know what they actually measure, which is not what the app implies.

What a sleep lab measures#

The reference standard for measuring sleep is polysomnography, the overnight lab study. It records brain waves through electrodes on the scalp, eye movements, muscle tone under the chin, airflow at the nose, breathing effort bands around chest and abdomen, oxygen levels, heart rhythm and leg movements. A trained technologist then scores the night in 30-second chunks into wake, light sleep (stages N1 and N2), deep slow-wave sleep (N3) and REM.

The key point: sleep stages are defined by brain activity. Deep sleep means large slow brain waves. REM means a particular brain and eye signature with the body's muscles switched off. There is no way to observe those directly from your wrist.

What your wearable measures#

A wrist or ring tracker records movement through an accelerometer, heart rate and heart rate variability through optical sensors, and, on many devices, skin temperature and an estimate of blood oxygen. From those signals, an algorithm infers whether you are asleep and guesses the stage, using patterns learned from lab data: heart rate slows and steadies in deep sleep, variability shifts in REM, movement suggests wake.

This is inference, not measurement. It works to a point because body signals do correlate with brain states. It fails where the correlation loosens: alcohol, beta-blocker heart medicines, fever, arrhythmias, an unusual physiology, or simply a loose band can all skew the guess. Two different brands on the same wrist on the same night routinely disagree with each other, never mind with the lab.

How accurate, honestly#

Validation studies comparing consumer wearables to polysomnography paint a consistent picture.

Two of these deserve emphasis. First, the wake problem: because trackers lean on stillness to infer sleep, a person lying motionless and awake at 3am is often scored asleep. Trackers are so least accurate in the very people most motivated to check them.

Second, the staging problem: for separating light, deep and REM sleep, agreement with the lab is modest, and a given night's deep sleep figure carries an error bar wide enough to make single-night readings close to meaningless. Newer algorithms improve year on year, and the best current devices are respectable at the group-average level. At the individual-night level, treat stages as a weather forecast, not a thermometer.

Orthosomnia: when the tracker becomes the problem#

Sleep clinicians coined the term orthosomnia for a pattern they began seeing once wearables spread: a preoccupation with achieving perfect tracked sleep that itself damages sleep.

It has a recognizable shape. The person checks the score first thing and lets it set the day's mood. A bad score produces effort, earlier bedtimes, longer lie-ins, canceled evenings, and effort is precisely what sleep punishes: lying in bed longer to farm a better score dilutes sleep and deepens the insomnia pattern. Some people trust the device over their own body, reporting broken sleep while feeling fine, because the app said so. And the nightly verdict adds performance anxiety at lights-out, raising arousal at exactly the wrong moment.

Sleep is one of the few health behaviors that worsens when monitored anxiously. If any of this is familiar, the highest-value intervention costs nothing: take the tracker off for two weeks, or at minimum stop opening the sleep tab, and rate your sleep by daytime energy instead. Many people sleep noticeably better within days, which tells you what was actually broken.

How to use tracker data well#

Used correctly, a tracker is a decent behavioral mirror. The rules:

Read trends, not nights. Weekly and monthly averages smooth out the staging noise. A single Tuesday means nothing. A month of drifting-later bedtimes means something.

Watch the two numbers trackers are actually good at. Schedule regularity, when you go to bed and when you get up, and approximate total duration. These are also, conveniently, the two levers with the most evidence behind them. A tracker that shows your weekend sleep timing swinging three hours from weekdays has found a real, fixable problem, no brain electrodes required.

Compare you to you. Baselines differ between people and devices. A sustained change from your own baseline, resting heart rate creeping up, sleep shortening across weeks, is more informative than any comparison to the app's population norms.

Use it for experiments, not verdicts. Two weeks of data with alcohol and two weeks without, caffeine cutoff at noon versus late afternoon, exercise timing shifted. Trackers shine at before-and-after comparisons because their errors are roughly consistent within one person.

Ignore the score and the stage bars. If you must look, look monthly.

For chronic insomnia specifically, prefer a paper sleep diary. CBT-I, the first-line treatment, is built on morning-after estimates, and clinicians increasingly ask patients to set the wearable aside during treatment because device numbers feed the exact monitoring the therapy is dismantling.

When the data hints at real sleep apnea#

Here is where trackers earn genuine clinical credit. Obstructive sleep apnea, repeated airway collapse in sleep, is common and heavily underdiagnosed, and some of its fingerprints show up in wearable data.

Patterns worth attention: overnight oxygen estimates that repeatedly dip in a sawtooth pattern or read persistently low across many nights. Some devices now explicitly flag possible breathing disturbances, and studies suggest repeated flags correlate meaningfully with true apnea. A high overnight resting heart rate or fragmented sleep graph every single night; and snoring captured by bedside or phone audio features.

The data alone proves nothing, wrist oxygen sensors are screening-grade and misread easily with movement or poor fit. What converts a tracker hint into a doctor visit is company: loud snoring, pauses in breathing witnessed by a partner, waking choking or gasping, morning headaches, unrefreshing sleep however long, daytime sleepiness, hard-to-control blood pressure.

Tracker flags plus any of those symptoms is a solid reason to ask about sleep apnea testing, which nowadays is often a simple home study. Bring the tracker data to the appointment. Used this way, it is a screening prompt doing exactly what screening should do.

At work#

In safety-critical jobs, drivers, operators, anyone at height, the tracker question I get is whether the data can prove someone is fit or unfit for duty. It cannot, and I push back when workplaces flirt with using sleep scores that way, both because the staging data cannot carry that weight and because sleep data is sensitive personal information.

Where trackers do help occupationally is self-management on rotating shifts: seeing in black and white that day sleeps after nights are running five broken hours is often the nudge that finally gets blackout curtains bought. The rule stands regardless of any score: a worker sleepy enough to nod off at the wheel or the controls needs assessment and rest, whatever the app says about last night.

What I actually see in clinic#

Two patients, same week, stay with me. The first: a well-rested man convinced by his stage graphs that he had a deep sleep disorder, sleeping eight hours, functioning perfectly, and anxious every morning. His only sleep problem was the graph. The second: a truck driver whose wife had been reporting his snoring for years, who finally came in because his watch kept flagging low overnight oxygen. He had significant sleep apnea, and treatment changed his mornings and, frankly, his safety on the road.

Same technology, opposite outcomes. The difference was never the accuracy of the device. It was whether the data was answering a question that mattered, or inventing one that did not.

The bottom line#

A sleep lab measures the sleeping brain. A wearable measures a wrist and guesses. The guess is respectable for when you slept and roughly how long, weak for stages, and meaningless as a nightly score, so read trends over weeks, compare yourself only to yourself, and let daytime energy outrank any number. If tracking makes you anxious, removing the tracker is treatment, not denial. And if the data keeps flagging disturbed breathing or low oxygen, especially alongside snoring or daytime sleepiness, do not argue with it or obey it. Take it to a doctor and ask about a sleep apnea test.