2019NEJM
Perez et al. (Apple Heart Study)
419,297 people
0.5% got an irregular-pulse alert. Among those who wore an ECG patch, 34% had atrial fibrillation, and 84% of alerts during the patch matched AF.
Find on PubMed ↗Research library · Tech & environment
Wearables and AI: what the data reveals
Watches, rings and AI models trained on millions of people are uncovering things doctors couldn't measure before: hidden heart rhythm problems, how many steps really matter, and the power of one-minute bursts of everyday activity.
Good evidenceUpdated Oct 2026 · 4 new studies7ksteps a day linked to 47% lower death risk; benefits keep rising more slowly after thatDing 2025, Lancet Public Health, 57 studies
3 burstsof 1-minute huffing-and-puffing activity a day tied to about 40% lower death risk in non-exercisersStamatakis 2022, Nat Med, 25,241 adults
84%of watch irregular-pulse alerts matched atrial fibrillation on a simultaneous ECG patchPerez 2019, NEJM, 419,297 people
01
The basics
What your wearable actually measures
Most wearables shine green light into your skin and measure how much bounces back as blood pulses, a method called photoplethysmography (PPG). That gives heart rate, heart rhythm, HRV and estimates of blood oxygen. A motion sensor (accelerometer) counts steps and movement. Some add an ECG, skin temperature or breathing rate.
Accuracy varies. Steps and resting heart rate are good. Sleep duration is fair; sleep stages are rough estimates. Calories burned are often far off.
The big change is AI. Models trained on hundreds of thousands of ECGs, retinal photos or wrist-sensor recordings now spot patterns people can't see, such as signs of past atrial fibrillation in a normal-looking ECG. The best use for you: track your own trends over weeks, not single numbers.
How a smartwatch reads your pulse
Green light shines into your wrist. Blood absorbs more light with each heartbeat, so the reflected light pulses. AI models turn that wave into heart rate, HRV and rhythm alerts.
02
How it works
From wrist to insight
1
Sensor measures
Light and motion sensors record your pulse and movement all day.
2
Raw signal
A wave of blood flow and a stream of movement data.
3
AI finds patterns
Models trained on huge datasets turn signals into heart rate, sleep and rhythm alerts.
4
You see trends
Week-by-week changes are the most useful part.
03
Measure it
How much to trust each number
- Steps and resting heart rate: reliable for tracking trends.
- HRV: good if measured at the same time each night or morning. Compare only with yourself.
- Sleep: total time is fair; light, deep and REM splits are estimates.
- Calories: treat as a rough guide. Lab tests found errors of 27–93%.
- Irregular rhythm alerts: take them seriously and get an ECG from a doctor to confirm.
Try your number
LowGetting thereStrong benefitExtra, smaller gains
100015000
Ding 2025: risk falls steeply up to about 7,000 steps, then more gradually.
04
What the research says
The key studies
2019Lancet
Attia et al.
180,922 patients
An AI model spotted people who had atrial fibrillation from an ECG taken during a normal rhythm, with 79% accuracy.
Find on PubMed ↗2022Nature Medicine
Stamatakis et al.
25,241 non-exercisers, wrist trackers
Three short bursts a day of vigorous everyday activity (stairs, fast walking, carrying shopping) were linked to about 40% lower death risk.
Find on PubMed ↗2025Lancet Public Health
Ding et al.
57 studies
7,000 steps a day, compared with 2,000, was linked to 47% lower death risk and lower risk of heart disease, dementia, depression and falls.
Find on PubMed ↗2023JAMA
Khurshid et al.
89,573 adults, wrist trackers
'Weekend warriors' who packed their activity into 1–2 days got similar heart benefits to those who spread it out.
Find on PubMed ↗2017Journal of Personalized Medicine
Shcherbina et al.
60 adults, 7 devices
Most wrist devices measured heart rate within about 5%, but none estimated calories well; errors ranged from 27% to 93%.
Find on PubMed ↗05
New in 2025–26
What the newest research adds
AI can now spot hidden heart disease from a standard ECG better than cardiologists, and smartwatches can pick up serious sleep apnoea. But the watch blood pressure alert misses most cases of high blood pressure, and glucose monitors have no proven benefit for people without diabetes.
The new studies, in order
Every study added to this page since January 2025. Green rings are the strongest evidence: large trials or meta-analyses.
What a watch blood pressure alert really tells you
Bress et al., 2026 (JAMA). Chance of having high blood pressure for adults aged 60+, based on Apple's own accuracy data. No alert still leaves a 1-in-3 chance.
NewJul 2025Nature
Poterucha et al. (EchoNext)
Trained on 1.2 million ECGs from 230,000 patients; tested in nearly 85,000
An AI reading ordinary ECGs found 77% of structural heart problems such as valve disease, compared with 64% for cardiologists. Of flagged patients who then had a heart scan, nearly three in four had heart disease.
Good evidenceRead the source ↗
NewFeb 2026JAMA
Bress et al.
Modelling for US adults using Apple's validation data
The Apple Watch high blood pressure alert would miss about 59% of people with undiagnosed hypertension and wrongly alert about 8% of people without it. No alert does not mean your blood pressure is fine, so a cuff check is still needed.
Good evidenceRead the source ↗
NewAug 2026J Clin Sleep Med
Alavi et al.
147 adults, sleep lab and home nights
A Samsung Galaxy Watch picked up 94% of moderate to severe sleep apnoea compared with a sleep lab test, but also gave some false alarms (67% specificity). It can prompt testing, not replace it.
Early evidenceRead the source ↗
NewAug 2026JAMA Internal Medicine
Dower et al.
Narrative review of the evidence
There is no evidence that continuous glucose monitors improve health or prevent disease in people without diabetes. Normal ups and downs in blood sugar can be over-read, leading to needless food rules and worry.
Good evidenceRead the source ↗
What this changes for you
Treat watch alerts as a nudge to get properly tested, not as a diagnosis, and treat 'no alert' as no information. Check blood pressure with a cuff. A glucose monitor isn't needed if you don't have diabetes or prediabetes.
06
Take action
Getting the most from your device
Use it as a coach for habits, not a judge of every day.
- 1
Watch weekly trends
Look at 7-day averages for resting heart rate, HRV, sleep and steps.
- 2
Act on rhythm alerts
An irregular-rhythm alert deserves a proper ECG with your doctor.
- 3
Add 1-minute bursts
Take stairs fast, walk briskly between meetings or carry the heavy bags.
- 4
Aim for 7,000+ steps
If you're at 3,000, adding 1,000 a day each month is a good pace.
- 5
Don't chase sleep scores
Worrying about the score can itself harm sleep. How you feel matters most.
- 6
Check privacy settings
Know who can see and sell your health data.
07
Myth or fact
Common myths, checked
✕ MythCalories burned on your watch are accurate.
✓ FactThey're often off by a quarter or more. Don't eat back 'exercise calories'.
✕ MythSleep stages are measured exactly.
✓ FactOnly a lab sleep study measures stages properly; wearables estimate them.
✕ MythMore data means better health.
✓ FactData helps only if it changes habits. Some people become anxious from tracking.
✕ Myth10,000 steps is a scientific target.
✓ FactIt began as a 1960s Japanese pedometer slogan. Benefits start well below it.
08
Questions
Frequently asked
Ring or watch?
Rings are good for sleep, HRV and temperature; watches add GPS, screens and ECG. Both measure trends well.
Should healthy people use a glucose monitor?
It can teach you how meals affect you, but isn't needed by most. See the blood sugar page.
Can a watch replace a doctor?
No, but it can tell you when to see one and give useful data to share.
09
Test yourself
True, partly true or myth?
Read each claim and pick an answer: the card flips to show the verdict and why. Not sure? Use “Just reveal”.
Card № 01True or myth?
Smartwatch AF alerts can detect atrial fibrillation in people who don't know they have it.
True
Smartwatch AF alerts can detect atrial fibrillation in people who don't know they have it.
In the Apple Heart Study (over 400,000 people), about a third of those alerted had AF confirmed on a patch.
Card № 02True or myth?
Smartwatches count calories burned accurately.
Partly true
Smartwatches count calories burned accurately.
Heart rate is fairly accurate, but calorie estimates can be off by 20–40% or more.
Card № 03True or myth?
Sleep trackers can tell exactly how much deep sleep you got.
Myth
Sleep trackers can tell exactly how much deep sleep you got.
They estimate stages from movement and heart rate; agreement with lab sleep tests for stages is moderate.
Card № 04True or myth?
Step counters are generally accurate for walking.
True
Step counters are generally accurate for walking.
Most phones and watches are within about 10% of actual steps during normal walking.
Card № 05True or myth?
Tracking trends over weeks is more useful than looking at one day's data.
True
Tracking trends over weeks is more useful than looking at one day's data.
Single readings are noisy; trends in resting heart rate, HRV and sleep reveal real changes.
Card № 06True or myth?
AI chatbots can give reliable medical advice.
Partly true
AI chatbots can give reliable medical advice.
They answer many textbook questions well but can be confidently wrong. Use them to prepare questions for a professional, not to replace one.
Card № 07True or myth?
If your wearable says you're fine, you don't need check-ups.
Myth
If your wearable says you're fine, you don't need check-ups.
Wearables miss many problems, like high cholesterol, early cancers or blood pressure (most don't measure it accurately).
Card № 08True or myth?
Continuous glucose monitors are proven to help people with diabetes.
True
Continuous glucose monitors are proven to help people with diabetes.
They improve control and cut low-sugar episodes; benefit for people without diabetes is less clear.
Card № 09True or myth?
Wearing a tracker makes you more active.
Partly true
Wearing a tracker makes you more active.
Trials show a modest increase in steps, around 1,000–2,000 a day, which can fade without goals.
Card № 10True or myth?
Constant checking of health data can increase anxiety for some people.
True
Constant checking of health data can increase anxiety for some people.
'Orthosomnia' describes people losing sleep worrying about sleep scores.
Nothing here yet. Try another filter.
10
Sources
References and related topics
- Perez et al. (Apple Heart Study) (2019). NEJM. 419,297 people. PubMed ↗
- Attia et al. (2019). Lancet. 180,922 patients. PubMed ↗
- Stamatakis et al. (2022). Nature Medicine. 25,241 non-exercisers, wrist trackers. PubMed ↗
- Ding et al. (2025). Lancet Public Health. 57 studies. PubMed ↗
- Khurshid et al. (2023). JAMA. 89,573 adults, wrist trackers. PubMed ↗
- Shcherbina et al. (2017). Journal of Personalized Medicine. 60 adults, 7 devices. PubMed ↗
- Poterucha et al. (EchoNext) (2025). Nature. Trained on 1.2 million ECGs from 230,000 patients; tested in nearly 85,000. Source ↗
- Bress et al. (2026). JAMA. Modelling for US adults using Apple's validation data. Source ↗
- Alavi et al. (2026). J Clin Sleep Med. 147 adults, sleep lab and home nights. Source ↗
- Dower et al. (2026). JAMA Internal Medicine. Narrative review of the evidence. Source ↗