Smartwatches can provide useful health information, but many of those numbers are algorithmic estimates rather than direct measurements of the body, and some outputs are more accurate than others, a new study finds.

Smartwatches are widely used to track sleep, steps, heart rate, and other health measures, but their increasingly complex metrics can be confusing.

Adam Lepley, assistant professor at the University of Michigan School of Kinesiology, and his team developed a framework to help users understand what smartwatches measure and estimate, and how to interpret the data responsibly.

The results appear in the journal Sensors.

“The most important takeaway is that not all smartwatch metrics should be interpreted the same way,” Lepley says.

“Some outputs are relatively close to what the device’s sensor actually detects, while many others are estimates generated by combining sensor signals with proprietary algorithms, user characteristics and other assumptions.

“People shouldn’t take these metrics at face value. In many cases, these devices are better suited to tracking trends over time, rather than as precise laboratory measurements.”

Smartwatches combine information from several types of sensors. Optical sensors use light to detect changes in blood flow at the wrist, while motion sensors, GPS, and other technologies track movement, location, and additional signals. Algorithms then translate those signals into user-facing metrics.

Takeaways:

Smartwatch data are most useful for tracking changes within the same person over time. A consistent shift in resting heart rate, sleep, or activity may be more meaningful than a single unusual reading.

Resting and steady state heart rate, step count, and outdoor pace are generally more reliable than complex estimates such as calories burned, sleep stages, body composition, hydration, and recovery.

Accuracy can also be affected by movement, watch fit, temperature, sweat, skin tone, tattoos, and body composition.

Results may not be comparable across brands because manufacturers use different sensors, definitions and algorithms.

The researchers conducted a narrative review using topic-focused searches of PubMed, SPORTDiscus, and Google Scholar through June 2026. They also examined reference lists, technical and regulatory documents, and professional guidance.

Source: University of Michigan