Where, how tight, how warm

Where you wear it matters more than what you bought.

Tattoos, hairy arms, cold hands, strap tightness, which wrist. What the studies actually found, what each brand says, and the free fixes that move your numbers tonight.

Heart rate error, moving

Same tracker

On the wrist bone

20.5%

3 finger widths up

7.3%

Against a Polar H10 chest strap, n=10. Vermunicht 2025.

Questions people ask

Things that help

Things that hurt

Primary sources

Things people ask about

Does this affect accuracy?

Tested, and here is what came back. Where a brand says something the studies never checked, the badge says so.

Yes, it matters

Tattoos

Not a small drift: over ink the sensor often returns nothing at all.

SpO2

Matters

36% dropped out

The number

The study: n=25, tattooed skin against clear skin on the same arm, optical sensor against a chest ECG. Resting error 22.9% over ink against 2.9% on clear skin, and 9 of 25 people had total dropout. Ink darkness and tattoo age did not predict failure. Navalta and Bunn 2025.
What the brands say: Apple says "some tattoos, can also impact heart rate sensor performance." Garmin says ink "can block the light from the sensor." Polar says to avoid placing the sensor on them. Samsung says to wear its ring on a finger without tattoos. Google is silent.
What to do: Move the sensor to clear skin: higher up the forearm, the other wrist, or an armband.

Watch

Ring

Source

Higher body fat

Three of four watches ran well high on calories, and the error grew as body fat rises. Skin tone made no difference.

Calories

Matters for calories

+69 kcal in 10 min

The number

The study: n=58 Hispanic adults, a 10-minute cycling test against a COSMED K5 metabolic system. Average overestimate: Garmin Forerunner 955 +69 kcal, Samsung Galaxy Watch 5 +57, Apple Watch Series 8 +22, Fitbit Sense 2 +3 but with 13% of its readings implausible (over 4.5 times the true value). Error rose with body fat on every brand, p<0.01; skin tone had no significant effect. One short cycling protocol, so treat the size as indicative. Kostrna 2026.
What the brands say: No manufacturer addresses body composition in its calorie accuracy guidance.
What to do: Read workout calories as a high estimate, and more so the higher your body fat. Keep your weight current in the app, and track the trend rather than eating back the number.

Watch

Source

Cold hands

Cold cuts the optical signal roughly in half, and the device drops readings rather than guessing.

HRV

SpO2

Matters

Signal -41%

The number

The study: Ice over the forearm cut the raw optical signal by 41%, n=21. Warming the wrist for 15 minutes took blood-oxygen error from 4.1 points to zero, independent of skin tone, n=46. In a 10 C chamber the average looked fine while the ability to track change collapsed, with one ring going from 0.78 to 0.32 on concordance.
What the brands say: Apple says cold exercise means "skin perfusion in your wrist might be too low." Google says "cold weather can hinder your device’s ability to accurately measure your heart rate during exercise." Samsung says "keep yourself warm." Oura says "cold fingers can cause signal problems." Garmin says to warm up for 5 to 10 minutes first.
What to do: Warm up before you trust a workout reading, and expect gaps on cold nights and winter runs.

Watch

Ring

Source

Which wrist

Heart rate and sleep do not care which wrist. Step counts do, by about 1,250 a day.

Sleep

Steps

No for HR and sleep

Yes for steps

+1,253 steps

The number

The study: Both wrists worn at once: heart rate differed by 0.37 bpm, n=16; sleep was null on group means across 65 nights, n=13; the dominant wrist logged 1,253 more steps a day, n=12. Telling the app the wrong wrist moves activity by 22 to 26%.
What the brands say: Oura and Samsung both recommend the non-dominant hand for their rings.
What to do: Pick one wrist, stay on it, and set it correctly in the app.

Watch

Source

Too tight or too loose

Loose fails at every intensity. Too tight loses part of the pulse wave. The right pressure is personal.

HRV

Matters

23 to 47% better

The number

The study: A custom wrist rig with a load cell, n=17: tuning the pressure per person beat a universal setting by 23 to 47%, and loose (12 mmHg) failed at every intensity. A second rig, n=27, showed that excess pressure flattens the waveform. No study has tested a consumer strap at graded notches.
What the brands say: Apple says "tighten your Apple Watch band for workouts, then loosen it a bit when you’re finished." Google says "snug but not constricting, as a tight band restricts blood flow." Polar says tighten for training and loosen after. Fitbit Air says a pinky finger should slide under the band.
What to do: One notch tighter for workouts, back off for sleep.

Watch

Source

Ring finger and fit

A ring turned 30 degrees loses most of its signal. Which finger is best has never been tested.

HRV

Matters: rotation

Untested: which finger

-7.86 dB at 30°

The number

The study: n=10, 432 signal sets: at 30 degrees from the optimal position signal-to-noise falls to -7.86 dB, and doubling the LED power cannot recover it. Left against right hand: reliability 94.8%, n=96. Finger choice and tightness: no study.
What the brands say: Oura says "we recommend your index finger," sensor bumps on the palm side, "snug, not tight," and size down if you are between sizes. Samsung says the indicator goes on the palm side and to wear the sizer for 24 hours. Ultrahuman says index, middle or ring finger.
What to do: Sensors on the palm side, snug enough that it cannot spin overnight.

Ring

Source

No, it does not

Hairy arms

The one study that graded arm hair found no difference in accuracy. Shaving has never been tested on its own.

Does not matter

p = 0.29

The number

The study: Cardiac rehab patients, n=30, four-point photographic hair scale, Fitbit against a chest ECG. Hair density did not differ between the patients whose readings were accurate and the patients whose readings were not. Shaving was bundled with cleaning the sensor and taping the watch down, and that bundle helped 3 of 10. Vermunicht 2025.
What the brands say: Samsung lists "body hair, dirt, or other objects" as obstructions. WHOOP says hairy arms are marked safe. Apple, Garmin, Google, Oura and Polar do not mention hair at all.
What to do: Nothing. If your readings are poor, fix position and tightness first, because shaving is unproven.

Watch

Source

Skin tone

Average error is the same across skin tones. The spread is wider, and the effect shows up as missing readings rather than wrong ones.

SpO2

Does not matter on average

Bias: no difference

The number

The study: Null in the best-powered studies: n=53 with a balanced Fitzpatrick sample, and n=28 measured with an objective colorimeter rather than a self-report scale. Pooled bias was null in every stratum, but the limits of agreement were 2.2 times wider in dark skin, and dark-skin participants supplied a disproportionate share of the missing data for 2 of the 3 devices tested. Calories too: skin tone had no significant effect on calorie error across four watches, n=58 (Kostrna 2026), while body fat did.
What the brands say: No manufacturer addresses skin tone on a wear page.
What to do: Check completeness, not just the average. Gaps in the graph are the symptom here, not a shifted number.

Watch

Ring

Source

Two devices at once

Four optical devices worn together did not interfere with each other.

Does not matter

0 missing values

The number

The study: n=16, two armbands and two watches worn at once against a chest strap: biases ran from -0.05 to +2.93 bpm with no missing data. All Polar devices, and the paper carries no funding statement.
What the brands say: No manufacturer addresses it.
What to do: Wear both if you want to compare them.

Watch

Ring

Source

Nobody has tested it

Sharing a bed

A partner raises your limb movements 21% in a sleep lab. Nobody has checked what that does to a tracker.

Sleep

Untested

+21% movement

The number

The study: 12 couples, lab sleep studies, 4 nights: 61.5 limb movements a night when sharing a bed against 50.9 when not. Bed-partner status is not reported in any consumer-device validation study. Predicted direction is more wake scored, and that is a prediction rather than a measurement.
What the brands say: No manufacturer addresses it.
What to do: Expect more awake minutes on shared nights, and compare like with like.

Watch

Ring

Source

Small or large wrists

Nobody has tested wrist size with current hardware.

Untested

No data

The number

The study: One 2019 cardiac-rehab study found that wrist circumference did not matter. The 2026 studies list it as an uncontrolled variable rather than testing it.
What the brands say: No manufacturer addresses it.
What to do: Focus on position and tightness, which are tested.

Watch

Source

An old device

Device age has never been studied as a factor, across 249 validation studies.

HRV

Sleep

Steps

Untested

0 of 249

The number

The study: The umbrella review of 249 validation studies and 430,465 participants does not analyse device age. Firmware changes do move results: one Fitbit algorithm update took sleep-staging accuracy from 71% to 77% on the same hardware.
What the brands say: No manufacturer addresses it.
What to do: Keep firmware current. That is the only age-related lever with evidence behind it.

Watch

Ring

Source

Only the brands have an answer

Sweat, lotion and sunscreen

Every brand says keep the sensor clean and dry. No study has ever tested lotion or sunscreen at a wearable site.

Brand guidance

0 studies

The number

The study: Lotion, sunscreen and moisturiser have zero peer-reviewed tests at a wearable site as of 2026. Sweat has one: an n=14 prototype study found 3 to 8% changes to the shape of the signal and heart-rate error under 0.5 bpm.
What the brands say: Garmin says "avoid wearing sunscreen, lotion, and insect repellent under the watch." Polar says "even a small amount of dirt on the sensor can reduce its performance." Samsung says to wipe the sensor. Oura says dirty LEDs cause gaps.
What to do: Wipe the sensor. Treat the lotion advice as a manufacturer instruction, not a research finding.

Watch

Ring

Source

Free 7-Day Trial

Fix the inputs, then trust the number.

These factors are averages. Kygo shows which ones are moving YOUR readings, using your own wearable and food data.

Works with

Oura

Apple

Fitbit

Garmin

WHOOP

Polar

Google

Health

Free fixes

Things that help.

Ranked by how much they moved the numbers. Most cost nothing.

Wear it higher up the forearm

Three finger widths above the wrist bone instead of one cut movement error from 20.5% to 7.3%.

One study, n=10 · Brands agree

Helps

20.5% to 7.3%

The number

The study: Vermunicht 2025, Fitbit Inspire 2 against a Polar H10 chest strap, n=10 healthy adults. Agreement with the strap rose from 0.59 to 0.92. One study of ten people, never replicated, so treat it as a strong lead rather than a law.
What the brands say: Google says "a finger’s width above your wrist bone," and its sleep page says 2 to 3. Apple says "above the wrist bone (towards your elbow, not your hand)." Polar says "at least a finger’s width." WHOOP says "an inch above your wrist bone."

Watch

Source

Snug for workouts, loosen after

The right strap pressure is worth more than any spec-sheet number.

HRV

Custom rig · Brands agree

Helps

Up to 47%

The number

The study: Tuning contact pressure per person beat a universal setting by 23 to 47% on a load-cell rig, n=17, and loose failed at every intensity. See "Too tight or too loose" above for the full study line.
What the brands say: Apple says to tighten the band for workouts and loosen it afterwards. Google and Polar say the same thing in their own words.

Watch

Source

Armband or chest strap for arm-heavy sport

The same sensor moved from wrist to upper arm cut the error range roughly fourfold.

Strong, replicated

Helps

4x tighter

The number

The study: Three identical WHOOP 4.0 units worn on wrist, forearm and upper arm at once, n=28: treadmill agreement half-width 11.5 bpm at the wrist against 2.7 bpm at the upper arm (Moghaddam 2026). Replicated with Polar hardware, n=16. Rowing, elliptical with arm levers, weights and interval bursts are where the wrist fails.
What to do: For anything that grips, swings or bursts, move the sensor off the wrist.

Watch

Source

Warm up before you trust the number

Fifteen minutes of warmth took a low-perfusion blood-oxygen error to zero.

SpO2

Strong for mechanism · Brands agree

Helps

4.1 points to 0

The number

The study: Local wrist warming for 15 minutes in low-perfusion patients, n=46, took bias from 4.1 points to zero, independent of skin tone. Cooling cuts the raw optical signal 41%. See "Cold hands" above.
What the brands say: Garmin says to warm up for 5 to 10 minutes and take a reading before you start.

Watch

Ring

Source

Ring sensors on the palm side, snug

A ring that spins overnight is losing signal no LED power recovers.

HRV

One study · Brands agree

Helps

30° is enough

The number

The study: At 30 degrees from the optimal position, signal-to-noise falls to -7.86 dB and doubling the light output cannot compensate, n=10, 432 signal sets. See "Ring finger and fit" above.
What the brands say: Oura says index finger, sensor bumps on the palm side, snug rather than tight. Samsung says the indicator goes on the palm side.

Ring

Source

Pocket it when you push a cart or stroller

Wrist trackers lose a fifth of your steps when your arm is not swinging.

Steps

Direction replicated, size disputed

Helps

1 in 5 steps

The number

The study: Cart and stroller pushing: the wrist undercounted by 19.8% in the peer-reviewed study and by up to 96.6% in a conference abstract, so the direction is solid and the size is not. In a pocket the same walk lost 6.4%.
What to do: Pocket the phone or the tracker for the shop run, or accept the undercount.

Watch

Source

One wrist, and tell the app which

Switching wrists or setting the wrong one moves activity by about a quarter.

Steps

Strong

Helps

±22 to 26%

The number

The study: Set for one wrist and worn on the other, activity was overestimated by 22.6% or underestimated by 25.9%, n=45. In a supervised study, 15.6% of participants wore it on the wrong wrist. The dominant wrist also logs about 1,250 more steps a day. See "Which wrist" above.
What to do: Check the handedness and wrist fields in your app match reality.

Watch

Source

Charge in the shower, not in bed

Nearly half of participants were missing data by night five.

Sleep

HRV

Strong for the missingness curve

Helps

47% by night 5

The number

The study: Garmin, n=299, five nights: missing data rose from 22% of participants on night one to 47% on night five, and 30% of nights were lost overall. Battery is the authors’ inferred cause rather than a measured one.
What the brands say: Apple documents that Low Power Mode turns background heart rate and blood oxygen off entirely.
What to do: Charge it during a daily gap you already have, so the night gets recorded at all.

Watch

Ring

Source

Watch out

Things that quietly hurt.

Not wrong readings so much as readings taken in conditions the sensor or the model was never built for.

Rowing, elliptical arms, swimming

Any sport that grips or submerges the wrist beats every wrist sensor tested.

Hurts

4% to 30%

The number

The study: Rowing gave 13.4% error at the wrist against 3.8% walking on the same device. On an elliptical with arm levers, no wrist device reached acceptable agreement, n=50. Swimming: Garmin Venu Sq 4.05% dry and 29.95% wet, n=10.
What to do: Use a chest strap or an armband for these, or read the session as a rough shape rather than a number.

Watch

Source

Calories from a weights session

Heart rate survives lifting. The calorie model does not.

Calories

Hurts

+116%

The number

The study: n=62 against indirect calorimetry: heart rate correlated 0.96 to 0.97 during resistance training while energy expenditure read 116% high. Lee 2026, from the tables rather than the abstract.
What to do: Halve it.

Watch

Ring

Source

Stairs and slow walking

Below about 4 km/h step error jumps from 7% to 40%. Stairs fail at every pace.

Steps

Hurts

40% vs 7%

The number

The study: n=258 across 21 devices: 40 plus or minus 40% error at slow speeds against 7 plus or minus 16% at normal pace. On stairs, neither Fitbit tested met the 10% threshold in any condition, n=8.
What to do: If you walk slowly, read the trend rather than the total, and do not expect stairs to be counted.

Watch

Source

Your first nights with a new device

Everyone sleeps worse on night one, at home as much as in a lab. Judge it after a week.

Sleep

Hurts

7 nights

The number

The study: First-night effect: sleep onset went from 20 to 14 minutes and total sleep rose 12 minutes by night two, n=45, with no home against lab difference, n=30. It takes about seven nights for a stable personal mean.
What to do: Ignore week one. Compare the second week against the third.

Watch

Ring

Source

Battery saver mode

Low power settings switch background heart rate and blood oxygen off, and the gap looks like non-wear.

SpO2

Sleep

Brand guidance

Hurts

Off, not worse

The number

The study: Apple documents that Low Power Mode turns off background heart rate, background blood oxygen and heart-rate notifications. No validation literature exists on this, so it is manufacturer documentation only.
What to do: If a night or a day is blank, check whether a power-saving mode was on before you blame the sensor.

Watch

Ring

Source

Heat

Heat hurt more than cold for every device that moved.

Hurts

9.6 to 20.8 bpm

The number

The study: Ten devices in a 36 C chamber against a chest ECG, n=45: Fitbit Inspire 3 error more than doubled, from 9.6 to 20.8 bpm, and one ring rose 72%. The top devices barely moved.
What to do: Treat a hot-weather workout number as a rough shape, and do not assume heat helps the signal.

Watch

Ring

Source

Keep exploring

Related tools.

More free, evidence-based tools to get the most out of your wearable.

Accuracy vs lab

Oura

Apple

Garmin

Fitbit

Most Accurate Wearable

See which wearable is most accurate across 9 health metrics, backed by peer-reviewed research.

Wearables · 17+ studies

Open

Sensor & software focus

Hardware & Software Differences

What actually makes Garmin, Whoop, Oura, Apple Watch and Fitbit different, sensor by sensor.

Wearables · 6 devices

Open

Daily step counts

Step Count Accuracy

Which wearable counts steps most accurately, ranked by 20+ peer-reviewed studies.

Wearables · 9 devices

Open

Common questions

Wearable accuracy, answered honestly.

What is the single biggest thing I can do to make my wearable more accurate?

Move it higher up the forearm and keep it snug during exercise. In the one study that tested position, three finger widths above the wrist bone cut movement error from 20.5% to 7.3%. Ten people, so a strong lead rather than a law, but every manufacturer’s own guidance points the same way.

Do hairy arms affect a wearable?

In the only study that graded arm hair, hair density made no difference to accuracy. Shaving was never tested on its own. Samsung lists hair as an obstruction; WHOOP says hairy arms are fine.

Do tattoos stop a wearable reading your heart rate?

Often, yes: 36% of people in the one study had the sensor return nothing over ink at rest. Ink darkness and tattoo age did not predict it. Move the sensor to clear skin.

Does skin tone affect wearable heart rate accuracy?

Not the average error, in the best-powered studies. The spread is wider and the effect appears as missing readings rather than wrong ones.

Does body fat affect smartwatch calorie accuracy?

Yes. Against a metabolic system during cycling, n=58, every watch tested overestimated calories and the error grew with body fat on every brand. Skin tone made no significant difference.

Does cold weather make my wearable less accurate?

Cold roughly halves the optical signal and devices drop readings. Warm up first; expect gaps in winter.

Does it matter which wrist I wear it on?

Not for heart rate or sleep. For steps, the dominant wrist logs about 1,250 more a day, so pick one and set it in the app.

Are smart rings more accurate than watches?

At night, slightly, for resting heart rate. By day under movement a ring placed 9th of 10 devices. A ring that rotates loses most of its signal, so fit matters more than form factor.

Can I trust the calories from a weights session?

No. Heart rate tracks fine, but energy expenditure read 116% high against a metabolic cart. Halve it.

Sources

Every figure on this page traces to a primary source below, with funding relationships, sample sizes and sign-convention traps carried alongside the number rather than hidden. Where a study contradicts its own abstract, we cite the table.

Hardware currency. No 2025 or 2026 flagship, including Apple Watch Series 11, Pixel Watch 5, Fitbit Air, WHOOP 5.0, Oura Ring 4 against a sleep lab, Galaxy Watch 9 and Galaxy Ring, has an independent peer-reviewed validation as of September 2026. Placement, pressure, cold and motion findings carry across generations. Model-specific error figures do not.

Fit, placement and wear site

Wrist position and heart-rate accuracy: Fitbit Inspire 2 vs Polar H10 ECG, n=10

Vermunicht 2025, Eur Heart J Digital Health

Fit, placement and wear site

Same sensor, three body sites: Whoop 4.0 on wrist, forearm and upper arm, n=28

Moghaddam 2026, Sensors

Fit, placement and wear site

Upper-arm vs wrist replication across nine activities, n=16

JMIR Cardio 2025

Fit, placement and wear site

Contact pressure and PPG signal quality: load-cell rig, n=17

Scardulla 2020, Sensors

Fit, placement and wear site

Wrist and fingertip PPG under graded clamp pressure, n=27

Scientific Data 2025

Fit, placement and wear site

Posture, arm height, age and skin tone on wrist PPG signal quality, n=1,142

Charlton 2025, PLOS Digital Health

Fit, placement and wear site

Ring rotation on the finger and PPG signal-to-noise, 432 signal sets per subject, n=10

Sensors 2025

Fit, placement and wear site

Dominant vs non-dominant wrist step counts, both wrists worn simultaneously, n=12

Park 2019

Fit, placement and wear site

Wearing the device on the wrong wrist, and how often people do it, n=45

Sensors 2019

Fit, placement and wear site

Wrist vs hip placement for sleep vs in-lab PSG, n=100

Zinkhan 2014, Sleep Medicine

Fit, placement and wear site

Five wear placements vs home PSG in children, n=137

IJBNPA 2024