Short answer: modern sleep trackers are reasonably accurate at telling you when you were asleep and roughly how long, and considerably less accurate at telling you which stage you were in. In laboratory comparisons against polysomnography — the gold-standard clinical sleep study — consumer wearables typically detect total sleep time within about 20 to 40 minutes, but classify individual sleep stages correctly only around 50 to 70 percent of the time (Chinoy et al., 2021, Sleep). That gap is the single most useful thing to understand about the device on your wrist.
It does not mean your tracker is useless. It means it is good at some jobs and bad at others, and knowing which is which changes how you should read your morning numbers.
How do sleep trackers actually decide you are asleep?
A clinical sleep study measures brain activity (EEG), eye movement, muscle tone, breathing, and oxygen. A wearable measures almost none of that. It has an accelerometer that senses movement, and usually an optical sensor that reads pulse and heart rate variability through the skin.
From those two streams, an algorithm infers sleep. The logic is simple at its core: sustained stillness plus a slowed, regular heart rate looks like sleep. Faster, more variable heart rate with micro-movements looks like REM. Very low heart rate and deep stillness looks like slow-wave sleep.
That inference works well in aggregate and gets shakier the more precise the question. Reviews of the field have made the same point repeatedly: wearables are useful for tracking sleep-wake patterns over time, but stage-level output should be treated as an estimate rather than a measurement (de Zambotti et al., 2019, Medicine & Science in Sports & Exercise).
What are sleep trackers genuinely good at?
Three things hold up well in validation studies.
Total sleep time. When researchers tested seven consumer devices against polysomnography, most estimated total sleep time with reasonable accuracy at the group level, though individual nights could be off by half an hour or more in either direction (Chinoy et al., 2021, Sleep). A later home-based test of four newer devices under unrestricted, real-world conditions found similar performance — good on sleep versus wake, weaker on staging (Chinoy et al., 2022, Nature and Science of Sleep).
Timing and consistency. Bedtime and wake time are the easiest things for a device to get right, because the transitions are usually obvious in movement data. This matters more than people expect, since regularity of timing predicts how you feel at least as strongly as raw duration does.
Trends over weeks. A single night’s estimate carries real error. Fourteen nights of estimates, averaged, carries much less. Direction of travel is the most trustworthy output any tracker produces.
Where do sleep trackers get it wrong?
Sleep stages. Deep sleep and REM percentages are the least reliable numbers on your screen. Devices have historically over-estimated light sleep and struggled to separate deep from REM, and different brands can report meaningfully different stage breakdowns for the same night. If you are anxious that you “only got 42 minutes of deep sleep,” the honest reading is that your device made an educated guess. Deep sleep is worth understanding, but not worth chasing on a per-night basis.
Wake after sleep onset. Trackers systematically miss brief awakenings. If you lie still while awake — which is exactly what happens during a 3 a.m. wake-up when you are trying not to disturb anyone — the algorithm often scores you as asleep. This is why waking at 3 a.m. can feel far more disruptive than your app suggests.
Fragmented nights. The more broken your sleep, the more the error compounds. This is a real problem for exactly the people who most want the data.
Bad nights in general. Accuracy tends to be highest in healthy sleepers with consolidated sleep, and lowest in people with disrupted sleep. The device is most confident when you need it least.
Why does this matter more for new parents?
Postpartum sleep is the hardest case for any algorithm. Sleep is short, broken into fragments, and interrupted by long periods of lying still while feeding, soothing, or simply waiting to see if the baby settles. Objective actigraphy studies of the first four postpartum months found that new mothers’ sleep is characterised less by dramatically reduced total time than by severe fragmentation — many brief awakenings scattered through the night (Montgomery-Downs et al., 2010, American Journal of Obstetrics and Gynecology).
A tracker will often report a total that looks almost fine while you feel demolished. Both things are true. Total time is not the variable that broke you; continuity is. If you are reading your data postpartum, the number worth watching is your longest unbroken stretch, not your nightly total, and certainly not your deep sleep percentage. Our guide to sleep debt for parents covers how to think about the accumulated cost without turning it into another thing to feel behind on.
There is also a practical point: a device that scores your still, awake, 3 a.m. feeding as “light sleep” is not lying to you maliciously. It simply cannot see the difference. Your own sense of the night is legitimate evidence.
Why does this matter in perimenopause?
Perimenopause creates a specific and well-documented mismatch between how sleep feels and how it measures. In the Wisconsin Sleep Cohort, women reported markedly worse sleep quality across the menopausal transition, yet objective laboratory measures did not show the same magnitude of decline (Young et al., 2003, Sleep). Reviews of sleep across the menopausal transition have consistently described this subjective-objective gap, alongside genuine increases in insomnia symptoms and night-time awakenings (Baker et al., 2018, Nature and Science of Sleep).
For a tracker, the practical consequence is this: hot flashes and night sweats produce brief arousals that you feel vividly and that the algorithm may partly smooth over. If your device says you slept 7 hours 10 minutes and you remember four separate episodes of throwing off the duvet, the device is not overruling your experience.
What a tracker can do well here is spot patterns you would not otherwise connect — a run of warmer, more restless nights clustering with a particular week, or a shift in overnight heart rate. Our posts on perimenopause and sleep and night sweats go deeper into the underlying mechanisms.
Can a tracker make your sleep worse?
It can, and there is a name for it. Clinicians described orthosomnia — a preoccupation with achieving perfect sleep data that itself drives sleep difficulty — after seeing patients whose distress centred on their tracker output rather than their symptoms (Baron et al., 2017, Journal of Clinical Sleep Medicine). The mechanism is straightforward: a poor score in the morning creates anxiety about tonight, anxiety raises arousal at bedtime, and higher arousal makes sleep harder.
The risk is highest when a device delivers a single verdict-style number with no explanation and no action attached. We have written separately about why sleep scores cause anxiety and what to watch instead.
How should you actually read your sleep data?
A few working rules that fit what the research supports:
- Trust the trend, not the night. Look at seven- to fourteen-day averages. Single-night numbers carry too much error to act on.
- Trust timing most. Bedtime, wake time, and consistency are the most accurate outputs and the most actionable.
- Treat stage percentages as texture, not truth. Useful for spotting a big change across weeks. Not useful for judging last night.
- Trust yourself over the device on continuity. If you remember being awake, you were awake, whatever the graph says.
- Compare a device only to itself. Cross-brand comparisons are close to meaningless because the algorithms differ. A change within one device over time is the meaningful signal.
- Use symptoms as the outcome. Daytime energy, mood, and focus are the real endpoints. Data is only useful when it explains them.
And one clinical note: if a tracker repeatedly flags low overnight oxygen, very high resting heart rate, or heavy fragmentation alongside daytime sleepiness or loud snoring, that is worth a conversation with a doctor rather than a settings change. Consumer devices are not diagnostic tools, but they can be a reasonable prompt. The signs of sleep apnea are worth knowing.
The calm version
Your sleep tracker is a decent estimator and a poor judge. It gets duration and timing broadly right, gets sleep stages roughly right at best, and misses the quiet awakenings that make a night feel hard — which is precisely why your own experience of the night still counts. Read the two-week trend, protect your wake time, and let the stage percentages be interesting rather than important. Mendtide is built around that principle: patterns and plain-language context instead of a nightly grade, with every claim traceable to research.
A tracker can tell you what your nights look like from the outside. It cannot tell you what they cost you. Only you know that, and that knowledge is not less valid for being unmeasured.