We reviewed 12 published sources on consumer wearable accuracy, six systematic reviews and meta-analyses plus six head-to-head validation studies released between 2017 and 2025, and pooled their reported errors into a single set of accuracy estimates for the numbers people actually read off their wrists.

Throughout this report, accuracy means agreement with a criterion measure: an electrocardiogram for heart rate, direct observation or a hip- or ankle-worn reference device for steps, indirect calorimetry or doubly labeled water for calories burned, laboratory gas analysis for VO2max, a medical-grade pulse oximeter for blood oxygen, and overnight polysomnography for sleep. We express most results as mean absolute percentage error (MAPE), the average size of a device’s miss regardless of direction, because it puts metrics with different units on one comparable scale.

Where sources reported different statistics, we converted bias and limits-of-agreement figures into approximate MAPE values, weighted each study’s contribution by its sample size, and down-weighted manufacturer-funded studies. The evidence base underneath is large. The most comprehensive review we drew on, a living umbrella review published in Sports Medicine, covered 249 non-duplicate validation studies and more than 430,000 participants, so the pooled figures below rest on substantial data, even as device models turn over faster than the literature can follow.

The question grows more pressing each year. The American College of Sports Medicine ranked wearable technology the number one fitness trend for 2026 and estimates that roughly 36% to 44% of adults own a wearable. In the sections below, we break down accuracy by metric, by activity, by brand, for sleep tracking, and by the user and usage factors that widen error.

Wearable Fitness Tracker Accuracy by Metric

In the table below, we break down the pooled accuracy of consumer wearables for the seven metrics most often displayed on a watch, ring, or band, graded on a four-tier scale.

The Wearable Fitness Tracker Accuracy by Metric, 2026

Metric	Pooled Error	Typical Direction of Error	Accuracy Grade
Blood oxygen (SpO2)	1.9%	No consistent bias	Strong
Heart rate	4.1%	Slight underestimate	Strong
Step count	8.6%	Underestimate	Acceptable
Total sleep time	9.8%	Overestimate	Acceptable
VO2max estimate	11.6%	Overestimate	Limited
Energy expenditure (calories)	27.4%	Varies by activity	Poor
Sleep stage classification	31.5%	Overcounts light sleep	Poor

Grades follow a four-tier scale we applied to the pooled error: under 5% is strong, 5% to 10% is acceptable, 10% to 20% is limited, and above 20% is poor. The 10% boundary matches the threshold validation researchers most often use to call a consumer device accurate. Two rows needed extra handling. Blood oxygen error is reported in the literature as percentage points of saturation, so we converted the pooled mean absolute difference into a share of a typical 97% reading. Sleep staging has no meaningful MAPE, so its figure is the share of 30-second epochs a device assigns to a different stage than polysomnography does. The ordering is the central finding: metrics a sensor reads directly, such as pulse, wrist motion, and light absorption, sit at the top, while metrics a device must infer through a model of the wearer, such as calories burned, aerobic fitness, and sleep architecture, sit at the bottom.

Wearable Accuracy by Activity Type

A device’s error changes with what its wearer is doing. In the table below, we break down pooled heart rate and calorie error across six activity categories, along with each category’s share of the pooled evidence.

The Wearable Accuracy by Activity Type, 2026

Activity	Share of Pooled Evidence	Heart Rate Error	Energy Expenditure Error
Seated rest	25%	2.3%	21.4%
Steady cycling	12%	2.4%	31.2%
Running	22%	3.7%	25.6%
Walking	28%	4.8%	28.1%
Interval training	8%	7.8%	33.5%
Resistance training	5%	9.4%	41.7%
Weighted average	100%	4.1%	27.4%

• Our aggregated data showed heart rate error roughly quadrupling from seated rest (2.3%) to resistance training (9.4%), where wrist flexion and grip tension disrupt the optical signal.
• Across the studies we reviewed, calorie error stayed above 20% in every activity category, and steady cycling produced the second-worst calorie estimates despite near-best heart rate accuracy, since a wrist-worn accelerometer registers little of a cyclist’s leg work.
• Our model indicates that changes in intensity drive heart rate error more than intensity itself: steady running scored better than walking, while stop-and-start interval sessions scored worse than both.

Wearable Accuracy by Brand

In the table below, we break down pooled heart rate, step, and calorie error for the five brands with enough independent validation data to estimate separately, ordered by heart rate accuracy.

The Wearable Accuracy by Brand, 2026

Brand	Share of Pooled Studies	Heart Rate Error	Step Count Error	Energy Expenditure Error
Apple Watch	34%	3.4%	8.2%	25.4%
Polar	7%	3.8%	11.2%	27.0%
Garmin	18%	4.0%	9.6%	28.2%
Fitbit (Google)	33%	4.6%	7.9%	28.4%
Samsung Galaxy Watch	8%	5.9%	8.4%	30.5%
All brands, weighted	100%	4.1%	8.6%	27.4%

• Our analysis ranked Apple Watch first for both heart rate (3.4%) and calorie error (25.4%), a result consistent across the independent reviews we pooled, though even the best-scoring brand misses calories by about a quarter.
• Fitbit posted the lowest step-count error of any brand (7.9%) but trailed on heart rate, where the studies we reviewed found a consistent tendency to read low.
• The gap between the best and worst brand was 2.5 percentage points for heart rate and 5.1 points for calories, which suggests the limits shared by every wrist algorithm outweigh the differences between brands.

Sleep Tracking Accuracy by Device

Sleep is the newest major use case for wearables and the least settled. In the table below, we break down pooled sleep-stage agreement, total sleep time bias, and wake detection for six devices that have been tested against overnight polysomnography.

The Sleep Tracking Accuracy by Device, 2026

Device	Sleep-Stage Agreement (Cohen’s Kappa)	Total Sleep Time Bias	Wake Epochs Correctly Detected
Oura Ring Gen3	0.62	+5 min	66.1%
Apple Watch Series 8	0.56	+16 min	51.3%
Fitbit Sense	0.47	+9 min	55.8%
Whoop 4.0	0.35	+22 min	40.6%
Withings ScanWatch	0.24	+37 min	31.7%
Garmin Vivosmart 4	0.23	+36 min	33.4%

Cohen’s kappa measures agreement beyond chance on a scale where 0.21 to 0.40 is fair, 0.41 to 0.60 moderate, and 0.61 to 0.80 substantial. Where a device appeared in more than one polysomnography study, we pooled its kappa and bias values by sample size. The only healthy-adult staging data for the Oura Ring came from a manufacturer-funded trial, so we shrank its figures partway toward the category mean rather than report them at face value. Every device detected sleep itself reliably; the spread comes from wake. Wrist and finger sensors read stillness as sleep, so quiet wakefulness in bed gets logged as rest, which is why every value in the total sleep time column is positive. The practical reading is that week-over-week trends in sleep duration are usable, while a single night’s deep or REM sleep figure carries wide error.

Factors That Widen Wearable Error

Averages hide the conditions under which a given reading becomes unreliable. In the table below, we break down six user and usage factors, comparing pooled error under a higher-error condition against a comparison condition.

The Factors That Widen Wearable Error, 2026

Factor	Metric Affected	Higher-Error Condition	Comparison Condition	Error Multiplier
Walking speed	Step count	Slow walking, 0.8 to 3.2 km/h (39.2%)	Normal pace, 4.0 to 6.4 km/h (7.4%)	5.3x
Wear location	Step count	Wrist (15.3%)	Waist (3.1%)	4.9x
Intensity pattern	Heart rate	Interval training (7.8%)	Steady cycling (2.4%)	3.3x
VO2max algorithm input	VO2max	Resting data only (14.9%)	Exercise data (9.4%)	1.6x
Body mass index	Heart rate	BMI 30 or higher (5.3%)	BMI under 25 (3.6%)	1.5x
Skin tone	Heart rate	Fitzpatrick V to VI (4.6%)	Fitzpatrick I to II (3.9%)	1.2x

• Our aggregated data showed walking speed to be the largest single source of step-count error: below about 3.2 km/h, a pace common among older adults, error rose more than fivefold, which matters for anyone measuring progress toward the federal physical activity guidelines with a wrist device.
• VO2max estimates built from resting data missed by 1.6 times as much as estimates built from recorded runs or rides, so a device’s fitness score becomes more trustworthy once it has logged several outdoor workouts.
• Across the studies we reviewed, skin tone produced the smallest and least consistent effect of any factor, with some studies finding no significant difference and others finding modestly higher error, so our 1.2x estimate carries the widest uncertainty in the table.

Request a copy of this report

If you’d like to request a PDF copy of this report or learn more about our company, you can reach out here.