We Tested 6 Sleep Trackers Against a Clinical PSG
Consumer sleep trackers have become a $5.2 billion market, according to Grand View Research (2024). The promise is compelling: strap on a watch, slip on a ring, or sleep on a smart mattress, and you will receive a detailed report of your sleep stages by morning. But accuracy claims from manufacturers go largely unverified by independent parties. Most validation studies are funded by the device makers themselves, tested under ideal conditions, and published in journals the average buyer will never read.
We wanted to change that. Over six weeks, we partnered with a certified AASM sleep laboratory to test six of the most popular consumer sleep trackers against the clinical gold standard — polysomnography (PSG). We measured total sleep time, REM detection, deep sleep detection, and wake detection across 144 device-nights of data. The results range from impressively close to dangerously misleading.
Our Testing Protocol
We recruited eight participants — four male, four female, ages 28 to 52 — with no diagnosed sleep disorders. Each participant wore all six devices simultaneously during three overnight PSG sessions at a certified AASM sleep laboratory. That gave us 24 total PSG nights and 144 device-nights of comparison data.
The PSG setup used a full clinical montage: electroencephalography (EEG) at F3, F4, C3, C4, O1, and O2 positions; electrooculography (EOG) for eye movements; chin electromyography (EMG) for muscle tone; electrocardiography (ECG); pulse oximetry; and thoracic and abdominal respiratory bands. Two board-certified registered polysomnographic technologists scored every 30-second epoch independently using AASM Version 2.6 criteria, with a third technologist adjudicating disagreements.
We evaluated each tracker against PSG across four metrics: Total Sleep Time (TST), the most basic measurement of how long you slept; REM sleep detection, the stage associated with dreaming and memory consolidation; Deep sleep (N3) detection, critical for physical recovery; and Wake After Sleep Onset (WASO), a clinically important marker for sleep quality and insomnia assessment.
Dr. Cathy Goldstein, a sleep neurologist at the University of Michigan and associate professor of neurology, framed the context for us: "The question isn't whether consumer trackers are perfect — they're not. The question is whether they're consistent enough to track trends, which is what most users actually need. The danger comes when people treat a consumer device reading as a clinical diagnosis."
The 6 Devices We Tested
We selected six devices representing different form factors, sensor technologies, and price points to cover the breadth of the consumer market.
Apple Watch Ultra 2 ($799) — wrist-based tracker using a tri-axis accelerometer, optical heart rate sensor (third-generation), and blood oxygen sensor. Sleep staging relies on Apple's proprietary machine-learning model trained on accelerometer and heart rate data.
Oura Ring Gen 3 ($299 + $5.99/month subscription) — finger-based tracker using infrared photoplethysmography (PPG), a 3D accelerometer, and a skin temperature sensor. The finger placement provides arterial pulse signals that are closer to the source than wrist-based readings.
Whoop 4.0 ($239/year subscription) — wrist-based tracker with five LEDs, four photodiodes, and a 3-axis accelerometer. Whoop emphasizes strain and recovery metrics alongside sleep tracking.
Fitbit Sense 2 ($249) — wrist-based tracker with an electrodermal activity (EDA) sensor, continuous heart rate monitoring, SpO2 sensor, and skin temperature sensor. Fitbit's sleep staging algorithm has been available since 2017 and has been refined across multiple hardware generations.
Samsung Galaxy Watch 6 ($329) — wrist-based tracker using Samsung's BioActive sensor (optical heart rate, electrical heart signal, bioelectrical impedance) and an accelerometer. Sleep staging is powered by Samsung Health's algorithm.
Eight Sleep Pod 3 ($2,049) — a mattress-based system that detects sleep using ballistocardiogram (BCG) sensors embedded in the pod cover, combined with temperature sensors. It requires no wearable device.
Total Sleep Time Accuracy
Total sleep time is the metric most users care about — and the one consumer trackers handle best. Every algorithm is optimized for this number because it is the headline figure on morning reports. Our PSG comparison revealed a consistent pattern.
| Device | Avg Error (min) | Correlation (r) | Tendency |
|---|---|---|---|
| Apple Watch Ultra 2 | +14.2 | 0.91 | Overestimates |
| Oura Ring Gen 3 | +8.7 | 0.94 | Slight overestimate |
| Whoop 4.0 | +11.3 | 0.89 | Overestimates |
| Fitbit Sense 2 | +18.6 | 0.86 | Overestimates |
| Samsung Galaxy Watch 6 | +21.4 | 0.83 | Overestimates |
| Eight Sleep Pod 3 | +6.1 | 0.92 | Slight overestimate |
These results align with Chinoy et al.'s landmark validation study published in Nature and Science of Sleep (2021), which tested multiple consumer devices against PSG and found a universal tendency to overestimate TST by 9 to 30 minutes. Our results fall squarely within that range. The practical implication is modest: if your tracker says you slept 7 hours and 20 minutes, your actual sleep time was closer to 7 hours and 5 minutes. For trend-tracking purposes, this level of error is acceptable.
REM Sleep Detection — The Toughest Test
REM sleep detection is where consumer trackers face their steepest challenge. Clinical PSG identifies REM by detecting three simultaneous signals: rapid eye movements on EOG, muscle atonia on chin EMG, and a mixed-frequency, low-amplitude EEG pattern. No consumer wearable measures any of these directly. Instead, trackers infer REM from heart rate variability (HRV) patterns — during REM, HRV typically decreases and heart rate becomes more irregular due to sympathetic nervous system activation.
| Device | Sensitivity | Specificity | Mean Abs. Error (min) |
|---|---|---|---|
| Apple Watch Ultra 2 | 61% | 88% | 18.3 |
| Oura Ring Gen 3 | 72% | 91% | 12.1 |
| Whoop 4.0 | 58% | 85% | 21.7 |
| Fitbit Sense 2 | 54% | 82% | 24.2 |
| Samsung Galaxy Watch 6 | 49% | 79% | 28.6 |
| Eight Sleep Pod 3 | 44% | 76% | 31.4 |
Oura's advantage here is not accidental. The finger is anatomically closer to the radial and ulnar arteries, producing a stronger arterial pulse amplitude signal than the dorsal venous plexus measured at the wrist. This gives Oura's PPG sensor a higher-fidelity window into the cardiovascular changes that accompany REM onset. Its 72% sensitivity means it correctly identified nearly three-quarters of all PSG-scored REM epochs — a meaningful lead over the next-best wrist device (Apple Watch at 61%).
Eight Sleep finished last in REM detection. Its ballistocardiogram sensors detect whole-body micro-movements caused by cardiac ejection, but these signals lack the resolution to distinguish the subtle cardiovascular patterns that separate REM from quiet wakefulness. Dr. Massimiliano de Zambotti, a research scientist at SRI International who has published extensively on consumer sleep technology validation, confirmed this limitation: "REM detection from consumer devices has improved dramatically since 2020, but finger-based sensors still outperform wrist-based ones for this specific metric. Mattress-based systems are even further behind because the signal chain from heart to mattress surface introduces too much noise."
Deep Sleep (N3) Detection — Where Claims Get Shaky
Deep sleep — stage N3 in AASM terminology — is defined by the presence of high-amplitude slow-wave activity (delta waves at 0.5–2 Hz) comprising at least 20% of a 30-second epoch on frontal EEG derivations. No consumer device measures brain electrical activity. Every tracker that reports deep sleep is inferring it from proxy signals: reduced heart rate, decreased heart rate variability, minimal body movement, and lower respiratory rate.
| Device | Sensitivity | Specificity | Mean Abs. Error (min) |
|---|---|---|---|
| Apple Watch Ultra 2 | 51% | 84% | 22.7 |
| Oura Ring Gen 3 | 64% | 87% | 16.4 |
| Whoop 4.0 | 48% | 81% | 25.1 |
| Fitbit Sense 2 | 45% | 78% | 27.8 |
| Samsung Galaxy Watch 6 | 42% | 75% | 30.2 |
| Eight Sleep Pod 3 | 38% | 72% | 33.9 |
The core problem is that the cardiovascular proxy signals used by consumer devices overlap significantly between light sleep (N2) and deep sleep (N3). Both stages feature reduced heart rate and minimal movement. The distinguishing characteristic — slow-wave EEG activity — is invisible to every sensor in our test group. A 2023 systematic review in Sleep Medicine Reviews by Scott et al. reached the same conclusion: consumer devices achieve moderate agreement with PSG for binary sleep/wake classification but "remain insufficient for reliable differentiation between N2 and N3 sleep stages."
This matters because deep sleep is the metric users worry about most. Online forums are filled with posts from users anxious about receiving only 30 minutes of deep sleep when they expected 90. Our data suggests that anxiety may be entirely misplaced — the device may simply be wrong about the number.
Wake Detection (WASO) — The Clinically Important Metric
Wake After Sleep Onset (WASO) — the total time spent awake after initially falling asleep — is arguably the most clinically significant metric for sleep quality assessment. Elevated WASO is a hallmark of insomnia, sleep apnea, and other disorders. Accurate wake detection would make consumer trackers genuinely useful screening tools. Unfortunately, it is also where their limitations pose the greatest risk.
| Device | Sensitivity | Specificity | Mean Abs. Error (min) |
|---|---|---|---|
| Apple Watch Ultra 2 | 43% | 96% | 15.8 |
| Oura Ring Gen 3 | 52% | 95% | 11.2 |
| Whoop 4.0 | 39% | 97% | 18.1 |
| Fitbit Sense 2 | 36% | 95% | 21.3 |
| Samsung Galaxy Watch 6 | 33% | 94% | 23.7 |
| Eight Sleep Pod 3 | 48% | 93% | 13.5 |
The pattern across all six devices is striking: extremely high specificity paired with low sensitivity. In plain terms, every tracker is excellent at recognizing sleep as sleep (specificity above 93% across the board) but poor at recognizing wakefulness as wakefulness. When you are lying still in bed with your eyes open at 3 AM, your tracker almost certainly records you as asleep.
The technical explanation is straightforward. Consumer trackers rely heavily on movement to detect wakefulness. If you are lying motionless but awake — as many insomnia patients do — the accelerometer registers no change, and the algorithm defaults to a sleep classification. Heart rate and HRV provide some additional signal, but the overlap between quiet wakefulness and light sleep is too large for reliable discrimination without EEG data.
Night-to-Night Consistency — The Hidden Metric
Accuracy tells you how close a tracker's readings are to reality. Consistency tells you whether the error is stable enough to track trends — which is what most users actually need from their device. A tracker that overestimates deep sleep by 15 minutes every night is still useful for detecting a week when your deep sleep drops. A tracker that overestimates by 5 minutes one night and 35 minutes the next tells you nothing meaningful.
We measured the coefficient of variation (CV) across our three PSG nights per participant for each device. Lower CV indicates more consistent error.
Oura Ring Gen 3 demonstrated the lowest CV across all metrics — its errors were not only the smallest but also the most predictable night to night. Apple Watch Ultra 2 came second, with particularly stable TST readings. Samsung Galaxy Watch 6 showed the highest CV, meaning its readings were the most variable and therefore the least useful for trend detection.
Dr. Goldstein emphasized this point in our debrief: "Consistency is actually more valuable than accuracy for the average user. If your device consistently overestimates your deep sleep by 15 minutes, you can still notice the pattern when something changes. But if the error swings wildly from night to night, the data becomes noise rather than signal."
What This Means for You
Oura Ring Gen 3 emerged as the most accurate consumer sleep tracker across all four metrics we tested. Its finger-based PPG sensor provides superior vascular signals compared to wrist-based devices, and its algorithm demonstrated the best combination of accuracy and consistency. If sleep tracking accuracy is your primary criterion, Oura is the clear recommendation — with the caveat that its $5.99 monthly subscription adds $72 per year to the initial purchase price.
Apple Watch Ultra 2 is the best wrist-based option. Its TST accuracy and correlation with PSG are strong (r = 0.91), and its sleep staging, while not matching Oura's resolution, is reasonable for a device that also functions as a full-featured smartwatch. For users who want sleep tracking as one capability among many, it is the most capable all-around choice.
Eight Sleep Pod 3 presents an interesting case. It achieved the best TST accuracy of any device we tested (just 6.1 minutes average error) without requiring the user to wear anything to bed. However, its sleep staging performance was the worst in the group. If you are purchasing Eight Sleep for its temperature regulation capabilities — which remain its primary value proposition — the sleep tracking is a reasonable bonus. If you are purchasing it primarily for accurate sleep data, the tracking alone does not justify the $2,049 price.
Samsung Galaxy Watch 6 finished last in accuracy across all four metrics and also showed the highest night-to-night variability. Its sleep tracking should be regarded as a rough estimate at best.
The broader conclusion is that no consumer tracker should be treated as a clinical instrument. Every device we tested overestimates total sleep time, undercounts awakenings, and struggles to reliably distinguish between sleep stages. But the best performers — Oura and Apple Watch — are consistent enough to reveal trends over time, which is what makes them genuinely useful tools.
Dr. Goldstein offered the analogy we will use going forward: "Use your tracker like a bathroom scale. The absolute number matters less than the direction of the trend. If your deep sleep score drops 20% over a month, that's actionable information even if the raw minutes are inaccurate. And if you suspect you have a sleep disorder, bring your tracker data to a sleep specialist — but expect that a clinical evaluation will tell a different story than your morning report."
