how your cortisol score works.
Your cortisol score estimates how far your body's strain signals (your heart at rest and while sitting, HRV, breathing and how long you sleep) sit from your own usual, using Apple Watch readings. It isn't a lab test, and it doesn't measure cortisol or any hormone.
What goes in
Six signals your watch already records:
- Resting heart rate
- Heart rate while you're sitting still
- HRV during the day
- HRV overnight
- How long you slept
- Your breathing rate overnight
To know when not to trust a reading, we also check wrist temperature, blood oxygen, workouts, mindful sessions and heart-rhythm records in Apple Health.
Most of the estimate comes from how your night and resting heart rate compare with your usual. Readings during the day can push it up or down, mostly when they're fresh and keep pointing the same way.
Compared with you, not with other people
We learn your usual for each signal from your last 4 weeks. If Apple Health already has a week of readings, you won't wait. Until then we lean on typical values for people like you, by age and sex.
A score of 40 means your signals are right at your usual. Higher means they've moved the way they tend to under strain: a higher heart rate at rest, lower HRV, a shorter night. Illness, alcohol and hard training move them too.
| Band | Score | What it means |
|---|---|---|
| Low | under 30 | Calmer than your usual |
| Usual | 30–49 | Around your usual |
| Raised | 50–59 | Higher than your usual |
| High | 60 and up | Well above your usual |
The bands describe distance from your own usual. They aren't a medical range.
Why it changes
Your night sets each morning's starting point. Readings during the day nudge it. When no new readings arrive, those nudges fade back toward the morning's starting point within a few hours. A single reading moves it only a little. An unusual stretch of sitting heart rate counts for less until more readings agree, so the score moves more when readings keep pointing the same way for a few hours.
Every point is accounted for
Tap your score to see how many points each signal adds or takes off. They add up to your number.
When we're less sure, we say so
A missing signal counts for less, and your estimate leans toward your usual instead of guessing high. You'll see "rough estimate" and what would help. Each estimate shows the time of its newest reading.
What throws it off
We set heart readings aside during and just after workouts and breathing sessions, so they don't read as stress.
AFib history from Apple Health pauses the score. The app doesn't ask about medical conditions or medicines, and the score doesn't adjust for things that change heart signals more than stress does: medicines such as beta-blockers or steroids, pregnancy and a pacemaker. If one of these applies to you, a higher or lower score may be them, not stress.
If your overnight readings suggest you may be run down, we dim the score. We don't yet adjust for alcohol or your cycle either, so a higher score after drinking or before your period may be them.
Don't use your score to guide any medicine or dose. Don't change a medicine because of it; talk to your doctor.
What we've tested
In public research recordings made with a different wristband (not Apple Watch), the daytime part of the score was higher during short lab stress tasks than at rest for most participants. That shows the signals respond to acute stress. It doesn't show how well the score tracks everyday stress. We haven't tested this score against cortisol.
Validation summary
These are the numbers behind that paragraph, for score model 1.0.4. AUC is a measure of how well a number separates two kinds of moment: 0.5 is a coin toss and 1 is perfect.
| Test | Data | Result |
|---|---|---|
| Lab stress tasks vs rest (pre-registered bar: median per-person AUC of at least 0.70) | 26 people, Empatica E4 wristband, scripted tasks of 0.5–7 minutes (Hongn et al. 2025) | Median per-person AUC 0.82 (mean 0.76, 95% CI 0.66–0.84). Met. |
| Same idea, a second dataset (heart rate only) | 20 people, fingertip pulse sensor (Birjandtalab et al. 2016) | Mean per-person AUC 0.68 |
| Ordinary stressful days vs calmer days | Simulated people, not real ones (an engineering check) | AUC around 0.6: weak, as the research predicts |
| A pre-registered comparison with our re-implementation of a simpler formula used by another app, based on population averages and given live heart rate | The same 26 people | Not met under the current model: the gain wasn't clearly above zero. We make no claim to be more accurate than other apps. |
What this means, and doesn't:
- Almost all of the lab signal came from heart rate while sitting. The wristband's HRV added little.
- The wristband isn't an Apple Watch, and short lab tasks aren't daily life. The overnight part of the score, the 4-week usual and the alerts have been checked only on simulated data.
- Samples are small (20–26 people per test), and the person who designed the method also ran the analysis.
- None of these tests involved measuring cortisol: no dataset we used had it.
Readiness (optional, off by default)
Readiness is an estimated number from 1 to 99 for how ready your body looks for effort today, compared with your own usual. The number stays hidden until you switch it on in Settings; the one daily movement suggestion in Tools uses it either way, and can only be made gentler by your cortisol score. It reads four overnight signals (resting heart rate, HRV while you sleep, how long you slept and your breathing rate overnight), your HRV across the last week and your workouts in Apple Health. Wrist temperature and blood oxygen are used only to spot a morning when you may be run down. It's set once each morning. Excellent is 85 and up, Good 67 to 84, Fair 50 to 66 and Low under 50. If Apple Health has AFib records from the last 30 days, readiness pauses.
Its real-data test failed. Before running it, we set a bar: readiness should pick out the mornings people said they felt least ready. On public data from Fitbit wearers who rated their readiness each morning, it didn't: no better than chance, and about what shuffled data scored. It also did no better than a version based on population averages. Recent training load was the one part that added a little. A second public dataset, with a different mood question, showed no signal either.
| Readiness test (pre-registered) | Data | Result |
|---|---|---|
| Picks out the least-ready mornings (bar: median per-person AUC of at least 0.60) | 12 people, Fitbit, morning readiness ratings (PMData) | Median per-person AUC 0.54 (95% CI 0.44–0.58), inside the shuffled-data range. Failed. Model 1.0.0 as frozen; 1.0.1 scored 0.53 on the same data |
| Personal usual beats population averages | Same people | Failed (no difference) |
| Training load adds information | Same people | Passed: a small gain (+0.02) |
| "Rested" mood ratings | 16 people, Fitbit (LifeSnaps) | AUC 0.44: no signal |
So treat readiness as one input next to how you feel, never a verdict. It doesn't measure fitness, and it can't tell you whether exercise is safe for you. If you answer the optional morning "How ready do you feel?" check-in, your answer stays on your iPhone; it isn't sent anywhere and doesn't change readiness or any score.
What it can't do
It can't diagnose anything, and it isn't intended for medical use. Check with a doctor in addition to using Cortisol IQ, and before making any medical decision.
Data: Hongn et al. 2025, "Wearable Device Dataset from Induced Stress and Structured Exercise Sessions", PhysioNet, doi:10.13026/he0v-tf17; Birjandtalab et al. 2016, "Non-EEG Dataset for Assessment of Neurological Status", PhysioNet, doi:10.13026/C26Q2Z (both ODC-By 1.0). Readiness: Thambawita et al. 2020, PMData, doi:10.1145/3339825.3394926; Yfantidis et al. 2023, LifeSnaps, doi:10.1038/s41597-022-01764-x (both CC BY 4.0).