How accurate is a sleep tracker at scoring rest stages

A consumer sleep tracker relies on sensors that detect movement and pulse characteristics rather than direct brain activity, meaning its nightly figures represent indirect estimates of your rest. Understanding how these tools perform requires looking at what their hardware detects, the laboratory standards used to evaluate them, and the systematic findings from independent validation research.

What wearable sleep technology senses

Consumer devices gather physiological signals to generate estimates of nightly rest. A 2019 review of wearable sleep technology describes consumer devices that collect signals such as heart rate and its variability, skin conductance and temperature in addition to movement, from which they claim to extract information about sleep ^[1]. A 2024 scoping review of 35 articles covering 62 wearable setups found a trend towards combining an accelerometer with photoplethysmography, the optical pulse sensor, and reported that devices using accelerometer data alone distinguished sleep from wake reasonably well but fell short at identifying multiple sleep stages, unlike setups that added the pulse signal ^[8].

The laboratory reference and human scoring

Validation studies evaluate commercial monitors against polysomnography, a comprehensive laboratory setup that records brain waves, eye movements and muscle tone. Even this benchmark involves human judgment. In a reliability programme with more than 2,500 scorers, 1,800 epochs and over 3.2 million scoring decisions, agreement with the majority score averaged 82.6 percent, was 67.4 percent for stage N3 and was lowest for stage N1 at 63.0 percent ^[10].

CIBDOL · The laboratory reference and human scoring

Detection of sleep versus wake periods

Across validation studies, wearable monitors consistently identify periods of sleep while struggling to recognise quiet wakefulness. In a sleep laboratory, 34 healthy young adults were recorded with polysomnography on three consecutive nights, one of them with deliberately disrupted sleep, each wearing research actigraphy and using a subset of the seven consumer devices tested, four wearable and three nonwearable ^[3]. Epoch by epoch, every device detected sleep with a sensitivity of at least 0.93, but specificity, the ability to detect wake, ranged from 0.18 to 0.54; sleep stage comparisons were mixed and the devices tended to perform worse on nights with poorer or disrupted sleep ^[3]. Most devices matched or exceeded research actigraphy on sleep and wake measures ^[3].

A study of 62 adults, 52 men and 10 women with a mean age of 46, who spent one night in a sleep laboratory wearing two to four of six wrist-worn devices, found that all devices detected more than 90 percent of sleep epochs but specificity for wake ranged from 29.39 to 52.15 percent, and Cohen's kappa ranged from 0.21 to 0.53, which the authors call fair to moderate agreement ^[5]. Most devices differed significantly from polysomnography on total sleep time, sleep efficiency, wake after sleep onset and light sleep, and the main epoch-by-epoch errors were wake, deep sleep and REM epochs that the devices scored as light sleep ^[5].

Agreement across specific sleep stages

When devices attempt to break rest into distinct phases, agreement with laboratory measurements drops. When 53 adults with a mean age of 25.4 spent one night in a sleep laboratory wearing six devices at once, agreement with polysomnography on whether each period was sleep or wake ranged from 86 to 89 percent, with Cohen's kappa from 0.30 to 0.51 ^[4]. Agreement on the specific stage or wake fell to between 50 and 65 percent, kappa 0.20 to 0.52 ^[4]. The authors conclude that all six devices were valid for field-based assessment of the timing and duration of sleep but all required improvement for specific sleep stages ^[4].

Hardware design also influences results. A multicentre study in Korea enrolled 75 participants at a tertiary hospital and a sleep clinic and compared 11 consumer trackers with in-lab polysomnography: 5 wearables and 6 devices that the authors call nearables and airables, 3 of each ^[6]. Across 349,114 epochs, agreement on sleep stage classification varied widely, with macro F1 scores ranging from 0.26 to 0.69 across the trackers ^[6]. Wearables showed a high proportional bias in sleep efficiency and nearables a high proportional bias in sleep latency ^[6].

Contactless hardware presents a different profile. A systematic review and meta-analysis of consumer sleep trackers that do not touch the body included 26 articles, 22 of them with data for pooling ^[9]. It found better accuracy in healthy participants using mattress-based devices with piezoelectric sensors, and reported that these contactless devices distinguished wake from sleep as well as actigraphy while also providing sleep stage estimates that actigraphy does not ^[9].

Pooled evidence and subjective experience

Broad reviews show systematic differences between device metrics and laboratory standards. A meta-analysis of 24 studies with 798 participants using wrist-worn consumer devices found significant differences between the devices and polysomnography in total sleep time, sleep efficiency, sleep latency and wake after sleep onset ^[7]. In subgroup analyses, one group of devices showed no significant difference from polysomnography in wake after sleep onset and the remaining devices none in sleep latency ^[7]. The authors conclude that the devices are not as reliable as polysomnography for measures such as total sleep time, sleep efficiency and sleep latency and that users should interpret results carefully, while noting that they can still be useful for tracking general sleep patterns ^[7].

A state-of-the-science review published in 2024 of wearables in sleep and circadian research states that the available evidence suggests consumer-grade devices exceed the performance of traditional actigraphy against polysomnography, but lists clear limitations: misclassifying wake during the sleep period, problems tracking sleep outside the main sleep bout or nighttime period, artefacts, and unclear performance in people with certain characteristics, noting that person-specific factors such as skin colour can reduce sensor performance ^[2]. The 2019 review of wearable sleep technology notes that little guidance exists on using these devices and names proprietary algorithms, device malfunction and firmware updates as critical factors to consider before a consumer tracker is used in clinical or sleep research protocols ^[1].

Recorded numbers also diverge from how people feel upon waking. A 2026 systematic review of five observational studies with 2,006 adults found poor to moderate agreement between consumer wrist-worn devices and validated subjective measures of sleep quality: device metrics explained only 2.5 to 16.2 percent of the variance in subjective ratings, device total sleep time correlated with same-day sleep diaries at r = 0.367, and the devices did not capture Pittsburgh Sleep Quality Index scores ^[11]. The authors conclude that device data should complement, not replace, validated subjective assessments ^[11]. Furthermore, the authors of a 2017 clinical report write that a growing number of patients are seeking treatment for sleep problems they have diagnosed themselves from periods of light or restless sleep shown by their trackers ^[12]. For these patients the tracker data often felt more consistent with their experience of sleep than validated methods such as polysomnography or actigraphy, and the authors write that the link these patients inferred between tracker data and daytime fatigue may become a perfectionistic quest for the ideal sleep ^[12].