What your wearable tells a health coach—and what it doesn't

What does a wearable actually provide?

A wearable provides observations made by a particular sensor, processed by particular software, at particular times. Some outputs are close to recorded events, such as a saved workout’s start time and duration. Others are estimates, such as energy expenditure, sleep stages, recovery, or a score built from several signals.

The distinction matters because a coaching app receives the output—not direct access to your body. Before the number can support useful coaching, it needs four pieces of information:

1. Source: Which device or app created it?
2. Signal: Was it measured, classified, calculated, or self-reported?
3. Comparison: Is the change unusual for this person under similar conditions?
4. Context: What happened that the sensor could not observe?

You could call this a source–signal–baseline–context check. It prevents a plausible-looking score from becoming a stronger claim than the data supports.

What can each common signal support?

Each signal answers a narrower question than its name may suggest.

Steps and energy expenditure

Step count can be a useful movement trend, especially when you keep the device and wearing pattern consistent. It still does not capture every kind of movement. Energy expenditure is a further calculation rather than a direct calorie measurement. A living umbrella review found that accuracy varies across metrics and devices, with wearables generally underestimating both steps and energy expenditure in the included reviews. (source)

A coach can therefore notice “movement has been lower than your recent pattern.” They should not convert that observation into “you burned exactly 430 fewer calories.”

Heart rate and HRV

Optical wearables often estimate heart rate reasonably well under controlled, resting conditions, but movement, fit, skin contact, device model, and the person wearing it affect performance. Secondary metrics such as HRV inherit those measurement concerns and add differences in sampling window and calculation. That review recommends emphasizing a person’s trend over universal cutoffs. (source)

Even the label “HRV” is not one universal number. Apple Health stores SDNN. (source) Consumer wearables may report other time-domain metrics such as RMSSD or turn several inputs into a proprietary score. (source) A coach should not compare those as if the units and methods were identical. For a step-by-step interpretation, read what a low HRV reading actually means.

Sleep

A wearable estimates sleep from the signals available to it; it does not reproduce a laboratory sleep study. A 2026 systematic review and meta-analysis found that performance varied substantially across devices. Wearables could support broad longitudinal tracking, but their limits prevented using them alone for diagnosis or detailed sleep-stage characterization. (source)

A responsible coach can use a repeated late bedtime as a conversation starter. They should not claim that 11 fewer estimated minutes of deep sleep explain a person’s fatigue. Our guide to wearable sleep tracker and sleep score accuracy explains the distinction in detail.

Workouts

A saved workout is stronger evidence that a recorded session occurred than an isolated movement estimate. It may include the workout type, start and end times, distance, heart rate, or active energy, depending on the source. It still cannot know the athlete’s intention, perceived effort, pain, technique, or reason for ending early.

That missing context can be short: “easy run,” “knee felt sore,” or “stopped because the meeting began.” Read how Apple Watch workout data becomes useful context for the full data flow.

Apple Watch wrist temperature

Apple Watch presents nighttime wrist temperature as a change from a personal baseline. It is not the same as an oral or core temperature. Apple explicitly states that its wrist-temperature feature is not a thermometer, cannot provide an on-demand measurement, and is not intended for medical diagnosis or treatment. Fit, sleep environment, physiology, and lifestyle factors can affect the result. (source)

A coach can say the Apple Watch trend changed and ask whether anything else changed. They should not say “you have a fever” or name an illness from that signal.

How should a coach turn signals into a next step?

The safest useful response gets no more specific than the evidence allows.

Consider this set of observations:

• HRV is below the person’s recent range.
• Estimated sleep duration is shorter.
• Resting heart rate is higher.
• A hard workout was recorded the previous evening.

Those signals support: “Several observations moved away from your recent pattern.” They may justify asking how the person feels, whether the measurements look complete, and whether the same pattern repeats.

They do not prove: “The workout caused poor recovery,” “you are getting sick,” or “you must skip training.” Those conclusions require information the sensors did not collect, and medical conclusions may require clinical evaluation.

A proportionate coaching sequence is:

1. Verify source and data quality.
2. Compare with the same person’s recent pattern.
3. State the change without naming a cause.
4. Ask for the smallest missing piece of context.
5. Suggest a reversible next step, or hand off to a qualified professional when symptoms or medical decisions are involved.

What control should the wearer keep?

Health data access should be specific and revocable. HealthKit requires apps to request access by data type, and people can change those permissions. Apple also prevents an app from reliably distinguishing “read permission denied” from “no readable data,” which is why an empty screen should not be described as proof that nothing was recorded. (source)

A coaching app should preserve the data source, explain what it can read, and avoid presenting imported observations as its own measurements.

How does Grio apply this boundary?

Grio is a general-wellness AI health coach. With permission, it uses read-only Apple Health context, including supported workouts and body signals, alongside information the user records in conversation. That the access is read-only is stated in our privacy policy. Its public App Store description explains that automatic records and user-provided context appear together, while Grio does not provide medical advice. (source)

The useful unit is not “the watch says 42.” It is “this device recorded this estimate, it differs from your recent pattern, and here is the context we still need.” That boundary lets automation reduce tedious tracking without pretending that a sensor knows the whole person.

What should you remember?

• Keep the device or app that created a signal visible.
• Separate recorded events from algorithmic estimates and scores.
• Compare the same metric with the same person’s pattern.
• Ask for intention, symptoms, and circumstances that a sensor cannot observe.
• Never turn one wearable number into a diagnosis, a cause, or exercise clearance.