Ear Beats

Why the AirPods Pro 3 heart rate accuracy numbers disagree

If you have read more than one review of the AirPods Pro 3 heart rate sensor, you have seen figures that do not line up. One write-up says the error is 1.67 percent. Another says 2.02 percent. A third says 1.23 percent. Elsewhere you may have seen gaps of 15 to 34 beats per minute, which sounds like a different device entirely.

Nobody has reconciled these. That is what this page does. Most of the disagreement is not disagreement at all. The tests measured different things against different references, and in two cases a number has been quoted in the wrong units or attributed to the wrong product. Every figure below is attributed to the publication that printed it, with its date.

The figures do not conflict. 2.02 percent is a 40-person lab study, 1.67 percent is a reviewer's two track runs, and the same study's limits of agreement are -10.27 to +10.22 BPM. Trust averages and trends, not a single reading or a zone boundary.

The two headline percentages measure different things

The 1.67 percent figure comes from CNET's track testing, reported by The Next Web on 20 June 2026. CNET put the AirPods Pro 3 at 1.67 percent average error and the Apple Watch Series 11 at 0.98 percent, both against a Polar H10 chest strap, on a four-lap track protocol. The Next Web notes CNET completed only two full AirPods runs.

The 2.02 percent figure is something else. It is the mean absolute percentage error from a peer reviewed study published in PLOS Digital Health on 30 April 2026 by Doherty and colleagues: 40 adults on a treadmill against a Polar H10, 16,735 paired five-second epochs. One number comes from a reviewer running laps twice, the other from a controlled lab protocol.

There is a further wrinkle behind much of the confusion. In a later CNET round, reported by iThinkDiff on 29 June 2026, the AirPods Pro 3 are listed at 1.23 percent average error and an average difference of 2.02 BPM, with the Apple Watch Series 11 at 0.63 percent and 0.89 BPM. So 2.02 circulates as a percentage in one source and as a BPM figure in another. If you have seen both quoted as percentages, one was mislabelled.

The 15 to 34 BPM gap is not an AirPods Pro 3 measurement

This one is worth correcting, because it does not belong to this product. The range appears in a SoundGuys article by Adam Birney dated 16 September 2025. SoundGuys attributes it to Lifehacker's testing of the Powerbeats Pro 2, not the AirPods Pro 3, and says so openly. It was written before AirPods Pro 3 data existed, assuming the two devices share a sensor.

That assumption does not appear to hold. Notebookcheck, also on 16 September 2025, reports that the AirPods Pro 3 uses an infrared PPG sensor pulsed at 256 times per second, while the Powerbeats Pro 2 uses a green LED sampling more than 100 times per second. Different light source, different sampling rate.

We could not read the original Lifehacker article, so we cannot confirm the 15 to 34 BPM range in the publication that produced it, and we are not repeating it as an AirPods figure.

What the peer reviewed study found

The PLOS Digital Health study of 30 April 2026 is the most detailed public measurement available. Participants moved through seated rest, supine rest, walking, and running or incline walking, at targets of roughly 40, 55, 65, 75 and 85 percent of age-predicted maximum heart rate. Rapid transitions between intensities were deliberately included.

On averages the results are strong. The mean bias carries a 95 percent confidence interval of -0.22 to 0.17, and by intensity, mean absolute error ran from 1.31 BPM at rest to 2.4 BPM during vigorous exercise.

A mean bias near zero means the sensor is not systematically reading high or low: over a whole session its average is close to the chest strap's. The authors say this absence of drift suggests the device stayed responsive to dynamic changes, though the proprietary signal processing stopped them testing rapid transitions directly.

MeasureValue
Mean bias-0.03 BPM
Mean absolute error2.08 BPM
Mean absolute percentage error2.02 percent
95 percent limits of agreement-10.27 to +10.22 BPM
Participants40 adults
Paired epochs16,735
Reference devicePolar H10 chest strap

Averages hide the number that matters to you

Here is what gets lost when a review prints one percentage. A low average error does not mean every reading is close. The same PLOS study reports a standard deviation of differences of 5.23 BPM and 95 percent limits of agreement of -10.27 to +10.22 BPM. So while the average error was about 2 BPM, individual five-second readings could sit roughly 10 beats either side of the chest strap value. The study describes greater dispersion at higher heart rates: the mean stayed accurate as workload rose, but the spread widened.

That is the honest reconciliation. A 2 percent error figure and a reading 10 beats off are both true descriptions of the same dataset. One is about the centre, the other about the spread.

The authors are direct about the consequence. They write that the observed dispersion constrains applications that require fine discrimination between adjacent training zones, and that the device should not be considered interchangeable with ECG-derived systems for tasks requiring beat-level precision.

Why optical sensors in the ear behave this way

Optical sensing shines light into tissue and reads how much is absorbed as blood pulses through. Anything that moves the sensor against the skin, or changes blood flow near the surface, degrades the signal. A systematic review of wrist-worn trackers published in Frontiers in Digital Health on 21 August 2026 by Hogue and colleagues reports that during high-intensity and high-intensity interval training protocols, device accuracy declined, with some devices experiencing data loss and increased bias, especially at heart rates above 160 BPM. It also states that motion artefact remains a major barrier during high-intensity or upper-body activities. That review covers wrist devices, not earbuds.

Apple's support documentation lists the practical versions. Poor skin contact inside the ear can cause a poor recording, and the ear tip giving the best seal may not be the one giving the best sensor contact. Cold weather can reduce blood flow in your ears. Earwax, moisture, skin conditions, piercings and metal jewellery can all interfere, and Apple says skin blood flow varies from person to person.

DC Rainmaker, in a review published 15 September 2025, found the sensor strong across indoor cycling, a steady 5K run and track intervals, while noting failures on an outdoor ride and that the signal drops out in water. That review publishes charts, not numbers.

What Apple publishes, and what it does not

Apple describes the hardware but says little about accuracy. The newsroom announcement of 9 September 2025 describes a custom photoplethysmography sensor shining invisible infrared light pulsed at 256 times per second. Footnote 5 of the AirPods Pro product page states that the monitor is intended for use during workouts with the Fitness app and compatible third-party apps on iPhone running iOS 26 and later. A workout framing, not a clinical one.

One honest caveat. Several secondary articles say Apple positions this as a wellness feature rather than a medical device. We looked for that language on Apple's own pages and found no medical-use disclaimer attached to the heart rate sensor. The FDA and medical-advice language on Apple's AirPods pages covers the Hearing Aid and Hearing Test features, not heart rate. So the accurate statement is narrower: Apple defines the monitor by its intended use in workouts, and publishes no claim that it is a medical or diagnostic device. This is not a medical device, there is no ECG and no arrhythmia detection, and nothing here is medical advice.

What this means for Ear Beats, plainly

Ear Beats displays what this sensor reports. It does not improve the sensor, so here is where the above limits what the app can tell you.

Trends and session averages are the strong case. With a mean bias of -0.03 BPM and a mean absolute error around 2 BPM in the PLOS data, an average or a trend over weeks is on solid ground. Resting readings are the best of the lot, at 1.31 BPM mean absolute error.

Single readings and zone boundaries are the weak case. With 95 percent limits of agreement of -10.27 to +10.22 BPM, a number on screen can sit about 10 beats either side of a chest strap. If a plan turns on whether you are at 149 or 161, Ear Beats cannot settle that, and the study authors say the dispersion constrains exactly this discrimination between adjacent zones. Use a chest strap.

Hard intervals, heavy upper-body work and anything that loosens the fit are where readings get worse. Cold weather and poor ear tip contact are documented failure modes in Apple's support notes, and water is a hard stop. None of this means distrusting the sensor for what it is good at. It means knowing which of the two you are reading.

Questions

So which number is right, 1.23, 1.67 or 2.02 percent?

All three were printed, and they are not measuring the same thing. 1.67 percent is CNET's earlier track round as reported by The Next Web on 20 June 2026, from only two full AirPods runs. 1.23 percent is CNET's later round as reported by iThinkDiff on 29 June 2026, alongside an average difference of 2.02 BPM. 2.02 percent is the mean absolute percentage error from the PLOS Digital Health study of 30 April 2026, with 40 adults and 16,735 paired epochs. The lab study has by far the most data behind it. Note that 2.02 appears as a percentage in one source and as a BPM figure in another, which is a common source of confusion.

Does the AirPods Pro 3 really read 15 to 34 BPM off during hard exercise?

We found no AirPods Pro 3 test reporting that. The range appears in SoundGuys on 16 September 2025, attributed to Lifehacker's testing of the Powerbeats Pro 2 and published before AirPods Pro 3 data existed. Notebookcheck reports the two use different sensors, infrared at 256 times per second on the AirPods versus a green LED on the Powerbeats. We could not read the original Lifehacker article, so we cannot confirm the figure in the publication that produced it, and it should not be quoted as an AirPods Pro 3 number.

How far off can a single reading be?

In the PLOS Digital Health study of 30 April 2026, the 95 percent limits of agreement against a Polar H10 chest strap were -10.27 to +10.22 BPM, with a standard deviation of differences of 5.23 BPM. So individual five-second readings sat roughly within 10 beats either side of the chest strap, even though the average error was about 2 BPM. The study notes greater dispersion at higher heart rates.

Can I use it for heart rate zone training?

Not for fine distinctions between neighbouring zones. The PLOS authors write that the observed dispersion constrains applications that require fine discrimination between adjacent training zones, and that the device is not interchangeable with ECG-derived systems where beat-level precision is needed. For a rough sense of effort and for session averages it is reasonable. Where the zone boundary itself matters, use a chest strap.

Why does my reading look wrong sometimes?

Apple's support documentation lists the known causes: poor contact between the sensor and the skin inside your ear, an ear tip chosen for the best seal rather than the best sensor contact, cold weather reducing blood flow in your ears, and earwax, moisture, skin conditions, ear piercings or metal jewellery interfering with contact. Apple also notes that skin blood flow varies significantly from person to person. DC Rainmaker reports the signal drops out immediately in water.

Is this a medical device, and can Ear Beats tell me if something is wrong?

No to both. This is not a medical device and Ear Beats is not a diagnostic tool. It displays what the sensor reports, and nothing on this page or in the app is medical advice. Apple's product page footnote states the built-in heart rate monitor is intended for use during workouts with the Fitness app and compatible third-party apps on iPhone running iOS 26 and later. There is no ECG and no arrhythmia detection. If you are worried about your heart, talk to a clinician.

Primary measurement source: Doherty C et al., PLOS Digital Health, 30 April 2026, doi:10.1371/journal.pdig.0001373 (40 adults, 16,735 paired 5-second epochs, Polar H10 reference, graded treadmill; bias -0.03 BPM, MAE 2.08 BPM, MAPE 2.02 percent, 95 percent limits of agreement -10.27 to +10.22 BPM, MAE 1.31 BPM at rest to 2.4 BPM vigorous). Background on optical sensing: Hogue T et al., Frontiers in Digital Health, 21 August 2026, doi:10.3389/fdgth.2026.1820224 (wrist devices, not earbuds). Apple sources: Apple Newsroom 9 September 2025 (infrared PPG pulsed 256 times per second); apple.com/airpods-pro footnote 5 (intended for use during workouts); Apple Support, Track your heart rate during workouts with AirPods Pro 3 (fit, cold weather, earwax, piercings, Apple Watch sensor fusion). Reviews: DC Rainmaker 15 September 2025 (charts only, no numeric metrics, no chest strap model named in the accuracy section); Notebookcheck 16 September 2025 (sensor hardware difference versus Powerbeats Pro 2); Macworld 15 September 2025 (reports DC Rainmaker, no own figures); SoundGuys 16 September 2025 (source of the 15 to 34 BPM range, attributed to Lifehacker Powerbeats Pro 2 testing). CNET figures are reported here only via The Next Web, 20 June 2026 (1.67 percent AirPods, 0.98 percent Apple Watch Series 11, two full AirPods runs) and iThinkDiff, 29 June 2026 (1.23 percent and 2.02 BPM AirPods, 0.63 percent and 0.89 BPM Apple Watch Series 11, four laps of a college track). Could not be pinned down: cnet.com and tomsguide.com block our user agent, so CNET's original article and both of its publication dates were not read, and Tom's Guide's chest strap model, workouts and figures are not reported here; the Lifehacker Powerbeats Pro 2 article behind the 15 to 34 BPM range was not located; whether CNET's 1.67 percent and 1.23 percent come from the same sessions re-analysed or separate testing is unknown; no source reports a proportion-of-readings-within-a-threshold figure; DC Rainmaker's session count is not stated. TechRepublic returned HTTP 403 and was not used for any figure. No figure on this page is an average or recalculation of numbers from different tests.

Heart rate zone calculator

Get Ear Beats on the App Store

More about Ear Beats ยท Contact us