A member of a GLP-1 program steps on a smart scale on Monday. Weight is down 0.9 kg, and the app says muscle mass is down 0.6 kg. One reading worries her about muscle; the next week’s, showing muscle up 0.4 kg, reassures her. Both are mostly noise.
The weight is real. A smart scale measures weight to within a few hundred grams [1]. The muscle and fat numbers are estimates, and the studies that test them against DEXA scans find they can be off by several kilograms [1][2]. That matters most for exactly the people now relying on them. Muscle loss is the side effect GLP-1 programs are built to manage, and the change they are watching for is smaller each month than the scale’s error.
What does a smart scale actually measure?
Only one number on a smart scale is measured directly: weight. Everything else comes from bioelectrical impedance analysis (BIA). The scale passes a small current between electrodes under your feet, and sometimes through handles, and measures how much the body resists it. Water and lean tissue conduct well and fat does not, so the scale feeds that resistance, your weight, height, age and sex into an equation to estimate body water, then lean mass, then fat [5].
Two consequences follow:
- The estimate is only as good as the equation fits you. Equations are built on reference populations. People with obesity, older adults and people losing weight fast are the hardest to fit.
- Anything that changes body water changes the reading. Hydration, a recent meal or drink, exercise and time of day all move the result, independent of how much muscle you have [5].
How accurate are smart scales for body fat and muscle?
Three independent studies give the clearest picture:
| Study | What was tested | What it found |
|---|---|---|
| Frija-Masson et al., 2021: about 50 hospital patients per scale, against DEXA [1] | Three consumer scales, including the Withings Body Cardio | Weight within a median of 0 to 0.3 kg. Fat mass off by a median of 2.2 to 4.4 kg. Muscle mass off by 4.0 to 6.6 kg, with the Withings Body Cardio at 4.0 kg |
| Bosy-Westphal et al., 2008: 106 adults, against DXA and MRI [2] | Three foot-to-foot devices and one tetrapolar device | Tetrapolar: no systematic bias, skeletal muscle within about ±4.7 kg. Foot-to-foot: systematic bias, skeletal muscle limits as wide as −9.5 to +3.9 kg |
| Siedler et al., 2023: 73 adults, against a four-compartment model [3] | 15 devices costing $24 to $349, including the Withings Body Cardio | Body fat errors from −3.5 to +11.7 percentage points depending on the model. Withings Body Cardio close on average but with a total error of 5.8 points on a single reading, and 1.9 points when tracking change |
The authors of the first study were blunt: “Smart scales are not accurate for body composition and should not replace DEXA in patient care” [1]. The 2023 study is more encouraging about change than about single readings: repeat measurements were highly consistent, and some models tracked change over 12 to 16 weeks reasonably well. But the authors still warned that “consumers should be wary” of their ability to track changes over time, and that it “largely depends upon the model in question” [3].
Precise is not the same as accurate
The 2023 study found something that explains why smart scale charts look so convincing. Repeat readings were extremely consistent: precision errors of 0.0 to 0.49%, and no device needed more than 1.36 percentage points of change to be distinguishable from measurement noise between back-to-back readings [3]. A smart scale will give you nearly the same wrong answer twice.
That consistency is real and useful, for one reason. A fixed bias, such as an equation that overestimates fat for your body type, cancels when you subtract one reading from another. That is why the same devices tracked change over 12 to 16 weeks with total errors of 1.9 to 2.9 percentage points, far better than their 3.3 to 12.6 point errors on single readings [3].
What does not cancel is anything that changes between readings and that the equation misreads as tissue. The main one is water. The scale estimates lean mass from body water [5], so a day of lower fluid intake looks like lean mass lost, and a salty dinner can look like muscle gained. Back-to-back precision says nothing about that, because back-to-back readings share the same hydration.
Why GLP-1s make the problem worse
GLP-1 drugs produce exactly the conditions under which BIA struggles: fast weight loss, shifting body water and a change in body proportions the equations were not fitted to. Early weight loss in particular includes glycogen, which the body stores with water, so the first weeks on treatment are when a scale that infers lean mass from water is most likely to misread what is happening. And the change programs want to detect is small.
In the STEP 1 trial, about two-fifths of the tissue lost on semaglutide was lean mass, by most calculations from the trial’s body composition substudy [4]. For a 100 kg adult who loses the trial’s average of about 15% over 68 weeks, that is roughly 6 kg of lean mass, or about 0.4 kg a month.
Now put that next to the scale. Because the scale measures weight accurately, an error in fat mass is mirrored by an equal error in lean mass. The Withings Body Cardio tracked change in body fat with a total error of about 1.9 percentage points [3], close to 1.9 kg for a 100 kg person. A steady loss of 0.4 kg a month takes about five months to clear that error. Against the error on a single reading, around 4 kg for the same scale against DEXA [1], it takes about ten months.
In other words, a weekly muscle number on a GLP-1 program’s dashboard is reporting the scale’s noise for most of the first half of treatment. A member who loses muscle and one who preserves it can look the same for months, and a single reading can show a loss that never happened.
What to track instead
Weight, as a trend. It is the one thing a smart scale measures well [1]. A weekly or rolling average removes day-to-day water swings and is a reliable signal of progress on treatment.
The behaviours that protect muscle. Resistance training and adequate protein are the standard countermeasures to lean mass loss on GLP-1s. Whether a member is actually doing strength sessions, staying active and eating enough protein can be measured from workout, activity and nutrition data far more reliably than the muscle loss itself. We covered the wider program design in how GLP-1s are reshaping digital health and why staying on GLP-1 pills is now the hard part.
A DEXA scan where it matters. For clinical programs, or members at higher risk such as older adults, a DEXA scan at the start and after several months is the only way to know how much lean mass has actually changed [1].
How products should show smart scale body composition
- Label it as an estimate. “Estimated body fat” is honest; a muscle number to one decimal place is not.
- Never show a weekly change. Show a four- to eight-week trend, and suppress changes smaller than the scale’s known error.
- Never trigger an alert from one reading. A sudden “muscle loss” warning is far more likely to be water than muscle.
- Keep conditions constant. Prompt members to weigh at the same time of day, before eating or drinking and after using the bathroom, on the same scale [5].
- Know which scale it is. Octapolar devices with handles measure the whole body and generally perform better than foot-to-foot scales [2][3]. Keep the source device with every reading so you can tell them apart.
One practical note for programs built on Withings scales: Withings accounts created since 13 October 2026 need a Withings+ subscription before their data reaches Apple Health, Health Connect or other apps. We covered what that means in Withings now charges new users to sync with Apple Health.
A smart scale is an excellent weight scale and a rough body composition estimator. For most people that difference barely matters. For a GLP-1 program, where the point is to catch muscle loss early, it decides whether the number on the dashboard means anything. Track weight as a trend, measure the behaviours that protect muscle, and save body composition decisions for a measurement that can actually see them.
References
- Frija-Masson, J., Mullaert, J., Vidal-Petiot, E., et al. (2021). Accuracy of Smart Scales on Weight and Body Composition: Observational Study. JMIR mHealth and uHealth, 9(4), e22487. https://pmc.ncbi.nlm.nih.gov/articles/PMC8122302/
- Bosy-Westphal, A., Later, W., Hitze, B., et al. (2008). Accuracy of Bioelectrical Impedance Consumer Devices for Measurement of Body Composition in Comparison to Whole Body Magnetic Resonance Imaging and Dual X-Ray Absorptiometry. Obesity Facts, 1(6), 319-324. https://karger.com/ofa/article/1/6/319/239580/Accuracy-of-Bioelectrical-Impedance-Consumer
- Siedler, M. R., et al. (2023). Assessing the reliability and cross-sectional and longitudinal validity of fifteen bioelectrical impedance analysis devices. British Journal of Nutrition. https://pmc.ncbi.nlm.nih.gov/articles/PMC10404482/
- Wilding, J. P. H., et al. (2021). Once-Weekly Semaglutide in Adults with Overweight or Obesity (STEP 1), including the body composition substudy. New England Journal of Medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC8089287/
- Kyle, U. G., et al. (2004). Bioelectrical impedance analysis, part II: utilization in clinical practice. Clinical Nutrition, 23(6), 1430-1453. https://doi.org/10.1016/j.clnu.2004.09.012