Your watch shows you a VO₂max figure. It updates quietly, it trends up or down, and it feels like a measurement. It isn’t. It’s an estimate produced by an algorithm that has never seen inside your lungs.
That matters, because VO₂max — your maximal oxygen uptake, the most oxygen your body can take in and use during intense exercise — is one of the strongest health markers we have. So it’s worth knowing how close the number on your wrist actually is, and what to do with it once you know.
Why the VO₂max number is worth getting right
An overview of meta-analyses covering more than 20.9 million observations from 199 cohort studies found that people with high cardiorespiratory fitness had roughly half the all-cause mortality risk of those with low fitness (hazard ratio 0.47), and that each 1-MET increase in fitness was associated with an 11–17% lower risk of dying from any cause (Lang et al., 2024). One MET is about 3.5 ml/kg/min of oxygen uptake — a small step on the scale, with a large effect attached to it.
The same review found the largest single association for incident heart failure: each 1-MET higher fitness came with an 18% lower risk. If a marker moves your risk that much, a 15% error in measuring it is not a rounding issue.
How does a fitness tracker estimate VO₂max?
No consumer wearable measures oxygen. The laboratory reference — cardiopulmonary exercise testing — puts a mask on your face and analyses the gas you breathe out while you work to exhaustion. Your watch has none of that. It infers.
Broadly, the algorithms fall into two families, and the distinction turns out to matter more than the brand on the case:
- Resting-based estimates use your demographics, resting heart rate and heart rate variability. No exercise required.
- Exercise-based estimates use the relationship between your heart rate and your actual work rate during a session — pace from GPS, or power from a meter.
Both are inferences. One is a much better inference than the other.
How accurate are fitness trackers at VO₂max?
The INTERLIVE network pooled 14 validation studies and found the split clearly. Devices using resting information overestimated VO₂max by 2.17 ml/kg/min on average, with limits of agreement stretching from −13.07 to +17.41 ml/kg/min. Devices using exercise-based information were near-perfect on average — a bias of just −0.09 ml/kg/min — but their limits of agreement still ran from −9.92 to +9.74 (Molina-Garcia et al., 2022).
Read those limits of agreement carefully, because they are the real finding. An average bias near zero means the errors cancel out across a group. It does not mean your number is right. Even with the better algorithms, an individual estimate can sit roughly 10 ml/kg/min either side of the truth — which, for a typical adult, is the difference between average fitness and genuinely good fitness. The authors’ own conclusion was that exercise-based estimation is fine at population level, but that individual error remains large enough that sport and clinical use still needs improvement.
A living umbrella review of 24 systematic reviews — 249 validation studies, 430,465 participants — reached the same shape of answer: wearables overestimated VO₂max by around 15% when working from resting tests and around 10% from exercise tests (Doherty et al., 2024). That review also noted something worth sitting with: only about 11% of consumer wearables released to date have been validated for even one biometric outcome, and the validation work done so far represents roughly 3.5% of what a comprehensive evaluation would require.
Device-specific studies fill in the picture. When the Apple Watch Series 7 was compared against a metabolic gas analyser in 19 adults, the laboratory mean was 45.88 ml/kg/min against the watch’s 41.37 — a mean absolute percentage error of 15.79% and an intraclass correlation of 0.47, which counts as poor reliability (Caserman et al., 2024). The Garmin fēnix 6 fared better in a comparable protocol, hitting 7.05% error and a concordance coefficient of 0.73 — clearing the study’s pre-set validation thresholds (Carrier et al., 2025).
So: not useless, not interchangeable, and not a measurement.
Why the error is worst at the extremes
There’s a pattern in the data that’s easy to miss. The Apple Watch study found the estimate tends to overestimate people with poor fitness and underestimate those with excellent fitness — and that accuracy improved in the fitter subgroup (error 14.59%, intraclass correlation 0.60) (Caserman et al., 2024).
This is what regression toward the population mean looks like in practice. An algorithm trained on a broad sample will pull unusual individuals back toward the middle. The consequence is uncomfortable: the two groups with most at stake — someone unfit who needs an honest baseline, and someone highly trained chasing a small gain — are the two the estimate serves least well.
What your tracker is genuinely good for
Accuracy and usefulness are not the same thing. A number can be systematically wrong and still be informative, provided it’s wrong in a consistent direction.
Your tracker’s real value is trend, not truth. If the algorithm, the device and the test conditions stay the same, a rise over eight weeks probably reflects a real change in your fitness — even if the absolute figure is several units off. What you should not do is compare your watch’s number against a friend’s different watch, against a published age chart, or against a laboratory result, and draw conclusions. Those comparisons are where the limits of agreement bite.
Treat it as a bathroom scale with an unknown offset. Useful for direction. Poor for absolutes.
How to get a VO₂max number you can act on
If you want the estimate to mean something, the evidence points to three things.
Prefer exercise-based estimates. The INTERLIVE meta-analysis is unambiguous that algorithms fed real work rate outperform those built on resting data (Molina-Garcia et al., 2022). A test where your actual power output is measured on a calibrated ergometer gives the algorithm far better raw material than a resting heart rate ever will. CAROL’s VO₂max test works this way — a submaximal ride at a personalised target power, with the work rate measured directly rather than inferred from GPS pace on a windy day.
Standardise the conditions. Same test, same device, same rough time of day. Trend data is only trend data if the method holds still.
Then change the input. A better estimate of your VO₂max doesn’t raise it. Training does — and the intensity matters more than the hours. In an eight-week workplace trial, reduced exertion high-intensity interval training (REHIT) — CAROL’s approach, built around two 20-second all-out sprints inside a roughly five-minute ride — improved cardiorespiratory fitness by 12%, against 7% for moderate-intensity continuous training (Cuddy et al., 2019). The sprint duration isn’t arbitrary either: cutting sprints from 20 seconds to 10 measurably attenuated the VO₂max gain (Nalçakan et al., 2018). Both were small studies, and individual responses vary — but the direction is consistent.
The bottom line
Your fitness tracker’s VO₂max is an estimate, and a noisy one. Exercise-based algorithms beat resting-based ones convincingly, yet even the good ones can land roughly 10 ml/kg/min from the truth for any given person. The error is largest exactly where you’d least want it — at the unfit and highly fit ends of the range.
Use the number for direction, not for diagnosis. Keep the test and the device constant so the trend means something. And remember that the estimate is the easy part: the marker that predicts how long you live responds to hard sprints, not to a better algorithm.
