A scale reports one number for several tissues. When someone loses 15 kg on a GLP-1, the question worth asking is what came off, and the honest answer is that the evidence for that question is much thinner than the evidence for the number itself. It is thin enough that both confident versions of the answer are wrong. Before reading anything on the semaglutide board as a body-composition claim, it is worth seeing what was actually measured.
The pivotal trials were built to weigh people
STEP 1 randomized 1,961 adults and set two coprimary endpoints: the percentage change in body weight and the proportion reaching a 5% reduction. Both are scale readings.[1] Body composition was assessed by dual-energy X-ray absorptiometry in a subgroup, not across the whole cohort — a design choice that is entirely reasonable for a weight-loss trial and entirely limiting for this question.
The same asymmetry runs through the program. Tens of thousands of randomized participants have been weighed. A far smaller number have been scanned. Anyone quoting a precise lean-mass percentage is quoting a smaller study than the one that produced the weight figure, which is the opposite of how those two claims are usually presented.
What the pooled body-composition data shows
A 2025 systematic review and meta-analysis gathered every study it could find measuring muscle mass on GLP-1 receptor agonists in adults with overweight or obesity. It came to 38 publications and 1,735 participants in total, and analyzed the diabetes and non-diabetes populations separately.[2]
In participants without type 2 diabetes, muscle mass measures fell by a mean of 1.41 kg (95% CI, −2.12 to −0.71) while fat mass fell by 6.02 kg (95% CI, −7.53 to −4.50). In participants with type 2 diabetes the muscle change was −0.74 kg and did not reach statistical significance (95% CI, −1.61 to 0.14, p = 0.10), against a fat-mass reduction of −3.18 kg.[2]
In both populations, the change in muscle measures accounted for less than 20% of the total weight reduction.[2] That is the single most useful sentence in the literature on this topic, and it is a pooled estimate from 1,735 people rather than a finding from any one trial.
Why the studies are hard to stack on each other
The review’s primary outcome was deliberately written as any measure used to estimate muscle mass, because the studies do not agree on one.[2] Some scan with dual-energy X-ray absorptiometry, some estimate from bioelectrical impedance, some read a blood marker. Those methods do not produce interchangeable kilograms, which is part of why the pooled confidence intervals are as wide as they are.
Duration is the second limit. Most of the scanned cohorts ran well under a year, so nothing here describes what several years of continuous treatment does to body composition. Continuous is how these drugs are sold, as the stopping article sets out.
Less than what, though
A figure like that means nothing without a comparator, and the obvious one is ordinary caloric restriction. The long-standing rule of thumb holds that about one quarter of weight lost by dieting is fat-free mass. A 2014 critical review of that rule found the relationship far less stable than the rule implies, and noted that lean-tissue losses with dieting tend to be small enough to raise questions about study design, statistical power and measurement reliability.[3]
So the comparison most people want — drug against diet, matched for weight lost — has not been run as a randomized trial. Under 20% on a GLP-1 sits below the quarter rule, and the quarter rule is itself contested. That is a defensible reason not to panic. It is not a finding that these drugs protect muscle, and no honest reading of the corpus produces one.
Mass and function are separate measurements
Every figure above is mass. What a reader actually cares about is whether they can carry shopping up stairs, and a DXA scan does not measure that. The 2025 review ends by calling for more detailed analysis of functional skeletal muscle, which is a plain statement that the functional evidence is not yet there.[2]
One trial does report function directly. STEP-HFpEF randomized 529 patients with obesity-related heart failure with preserved ejection fraction. Six-minute walk distance improved by 21.5 m on semaglutide against 1.2 m on placebo, an estimated difference of 20.3 m (95% CI, 8.6 to 32.1), alongside a mean weight change of −13.3% against −2.6%.[4] Physical capacity rose in the group that lost the most weight.
That is a real result and a narrow one. Those patients were selected for a specific cardiac condition in which excess weight itself limits exertion, so the finding travels poorly to a healthy 34-year-old. It also says nothing about tolerability over a long course, which the side-effect article covers separately.
What nobody has tested
The protocols people are told to follow alongside these drugs — a protein floor, resistance training twice a week, slower titration — are reasonable extrapolations from general exercise physiology. None of them was a randomized arm in the trials above. A seller presenting a protein target as clinically validated for GLP-1 users is describing something the evidence does not contain, and that is worth noticing when comparing what a program includes on the price boards. The same gap applies to the other molecule, priced on the tirzepatide board.
One further limit is worth stating plainly. Every scan above was performed on branded product at labeled doses, and whether a compounded preparation behaves the same way has not been established, as the compounding article sets out. The way this site establishes a figure before publishing it is described in the methodology.