Does a wearable read heart rate worse on darker skin? The honest answer is uneven

Consumer smartwatches and rings measure heart rate with photoplethysmography (PPG): they shine light, usually green, into the skin and read the pulse in the light that bounces back. Because melanin absorbs green light, the reasonable worry is that the reading degrades on darker skin — and the independent evidence says the effect is real but small, inconsistent, and smaller than the error introduced by the choice of device and by body movement, against a backdrop of validation studies that have barely tested the people most affected [s1][s2][s5].

Where the concern comes from

The physics is not in dispute. Melanin in darker skin absorbs more of the green light that most wrist sensors emit, leaving a weaker signal for the device to work with [s3]. The question is whether that translates into a measurable accuracy penalty in practice, and here the studies diverge rather than converge.

The most-cited early attempt to settle it was a 2020 analysis in npj Digital Medicine that tested consumer- and research-grade wearables across skin tones and conditions [s5]. It found no statistically significant difference in accuracy across skin tones — but it did find significant differences between devices and between activities, with absolute error during activity on average 30% higher than at rest [s5]. The authors' framing has held up: the biggest sources of error in a wrist heart-rate reading are which device you wear and whether you are moving, not the colour of your skin [s5].

The newer, more granular tests

More recent work adds nuance rather than overturning that. A 2026 study in Sensors put four wrist devices — Apple, Fitbit, Samsung and Garmin — on 58 Hispanic adults with Fitzpatrick skin types III to V during a cycling protocol alternating moderate (64-76% of maximum heart rate) and vigorous (77-95%) intensity, against a Polar H10 ECG chest strap as the reference [s2]. Every device showed small but systematic deviations from the ECG criterion (p < 0.001); Apple and Garmin had the lowest mean absolute error, while Fitbit and Samsung were worse [s2]. Crucially, the error spread widened among participants with higher body-mass index and the darkest (Fitzpatrick V) skin tones — what the authors call phenotype-linked agreement drift [s2].

A 2026 machine-learning study in Physiological Measurement built a multi-wavelength wrist dataset specifically to probe fairness, and reported that baseline error was higher for participants with darker skin tones and for women [s1]. But it also showed the gap was tractable: filtering out the device's least-confident predictions reduced the skin-tone parity gap from 1.2 beats per minute to zero [s1]. That points to the real problem being calibration and signal quality under poor conditions, not an immovable optical ceiling.

A 2026 narrative review in Cureus, which screened studies from May 2017 to May 2025 and included 23 of them, captured how wide the device-to-device range is [s3]. One 60-participant validation found a major smartwatch brand differed by under 5 beats per minute across skin tones (95% CI -3.2 to +4.1), while other brands underestimated heart rate by 10-15 beats per minute at rest and by more than 20% during vigorous activity in darker-skinned users (n = 75, p < 0.01) [s3]. Large atrial-fibrillation screening studies enrolling more than 400,000 people reported over 95% sensitivity, the review noted — but did not stratify their results by skin tone at all [s3].

The gap underneath the gap

The thread running through all of this is that the evidence base is too thin in exactly the place it matters. A 2026 scoping review in JMIR mHealth and uHealth mapped who actually appears in PPG validation studies, screening 734 papers and including 186 [s4]. Skin tone was reported in just 35 of them — 19% [s4]. The median share of participants with the darkest skin (Fitzpatrick types V and VI) in a given study was 0% [s4]. Older and larger bodies fared little better: pooled across all the studies, adults over 65 made up 20% of participants, people with obesity 14%, and underweight people 7% [s4].

In other words, the devices are most heavily validated on the populations least likely to expose a bias, which means published accuracy figures may flatter real-world performance for everyone the studies skipped [s4]. That is the same structural problem documented for pulse oximeters, whose oxygen readings run high on darker skin, and it is why a wrist heart-rate number cannot simply be assumed to travel unchanged across users.

What it means

For a resting heart rate glanced at over morning coffee, the skin-tone effect is unlikely to matter much; the devices agree well with a chest strap at rest [s5]. The caution applies where the number feeds a decision — an arrhythmia alert, a hypertension estimate, a training zone — and where the wearer is moving, because movement is where error balloons for every skin tone [s2][s5]. The measured takeaway is not that wearables fail on darker skin, but that their accuracy is device-specific, degrades with motion, and has been tested on too narrow a slice of people to promise equal performance to all of them [s1][s4].

What to watch is whether regulators and device makers adopt validation protocols that report results stratified by Fitzpatrick skin type and activity level, the fix the reviews keep calling for [s3][s4]. Until they do, the fairest description of a consumer heart-rate reading is that its error bars are wider, and less evenly measured, than the single number on the screen suggests.