Training · GLP-1s
A large literature with a specific hole in the middle of it
There is a lot of research on these drugs and muscle. Body composition substudies from the major trials, four systematic reviews in 2026 alone, randomised trials of drugs built to protect muscle during weight loss, and one landmark trial that added exercise to a GLP-1 and followed people for two years. Anyone who says the field is empty has not looked.
The hole is narrower and more specific than that. No published randomised trial has put progressive resistance training together with semaglutide, tirzepatide or retatrutide and measured what happened. The advice everybody gives, lift and eat protein while you are on it, is the one combination nobody has tested yet.
What the research contains
randomised
Adding exercise to the drug changed body composition, but not the weight on the scale
Measured in
195 adults with obesity and without diabetes, randomised after an 8 week low-calorie diet to exercise plus placebo, liraglutide 3.0 mg alone, both combined, or neither, for one year
Against liraglutide alone, adding exercise cut body fat percentage by a further 1.9 points (p = 0.009), roughly doubling the fat reduction of either treatment by itself. Only the combination improved HbA1c, insulin sensitivity and cardiorespiratory fitness. This is the trial the whole field rests on.
HumanHealthy weight loss maintenance with exercise, liraglutide, or both combined (2021)What this does not show
That the combination beat the drug for weight. It did not: the difference was 2.7 kg with a confidence interval from minus 6.3 to plus 0.8, which does not exclude zero. More importantly for anybody reading this as a training guide, the exercise was supervised cycling and circuit work, not progressive resistance training, and no protein target was set during the maintenance year. The thing usually recommended is not the thing this trial tested.
randomised
A drug added 6% lean body mass and improved no measure of physical function
Measured in
180 community-dwelling adults aged 70 and over with sarcopenia by gait speed and muscle criteria, across 38 sites in 13 countries, given bimagrumab or placebo monthly for 6 months alongside adequate diet and home exercise
Bimagrumab increased lean body mass by 7% against 1% on placebo, a 6 point difference with p below 0.001. Every functional endpoint was null: the Short Physical Performance Battery p = 0.13, six minute walk distance p = 0.16, gait speed p = 0.16. Function improved in both arms, and the authors attribute that to the diet and exercise both groups received.
HumanBimagrumab vs optimized standard of care for treatment of sarcopenia in community-dwelling older adults: a randomized clinical trial (2020)What this does not show
That muscle does not matter. It shows that lean mass on a scan and the ability to do something are different targets, and that adding the first does not deliver the second. This is the cleanest evidence there is that anybody chasing a body composition number, whether by drug or otherwise, has chosen the wrong endpoint.
randomised
The one trial of tirzepatide plus training has been retracted
Measured in
61 men with obesity and prediabetes, randomised to six groups across placebo, tirzepatide at 2.5 or 5 mg, and combined resistance plus aerobic training with or without each dose, over 6 weeks
It reported that training alone beat tirzepatide on fitness, strength and insulin, and that adding the drug to training gave no significant advantage over training alone. It was retracted in 2025, and PubMed now flags the original as a retracted publication.
HumanTirzepatide and exercise training in obesity (RETRACTED) (2024)What this does not show
Anything, and that is why it is here. Its numbers still circulate and its conclusion is exactly the kind people repeat, so anyone researching tirzepatide and training will meet it. It was also six weeks, in 61 men, split six ways, which is around ten people per group before anybody questioned the data.
observational
Four meta-analyses published in 2026 reach four different conclusions
Measured in
Four separate systematic reviews of body composition during incretin therapy, published within months of each other, covering overlapping but differently selected trial sets
One, restricted to trials reaching at least 10% weight loss, put fat-free mass at 33.3% of weight lost on incretins against 22.3% on diet alone. A second, four times larger and without that restriction, found no significant difference between incretins and lifestyle at all (p = 0.42). A third concluded lean mass was largely preserved and the losses modest. A fourth found the comparative evidence so heterogeneous and imprecise that it graded the certainty low to very low and declined to draw a conclusion.
ReviewBody composition remodelling during GLP-1-based therapy: a systematic review and meta-analysis using a hierarchical physiological framework (2026)What this does not show
That any of them is wrong. They differ because they selected different trials, used different comparators and set different weight-loss thresholds, which is what happens when an urgent question outruns the evidence. The fourth is probably the most honest, and it states that lean mass loss should not be read as sarcopenia while the evidence is also insufficient to rule out clinically important muscle loss in vulnerable people.
Why four reviews of the same question disagree
It is worth understanding, because it will keep happening and because it explains why two people can both cite a 2026 meta-analysis and say opposite things.
The choices that drive the disagreement are all made before any data is pooled. Whether to include only trials where people lost a lot of weight changes the comparison completely, because a lifestyle arm that lost three kilos is not a fair test against a drug arm that lost twenty. Whether the comparator is placebo, diet, or diet plus training changes the answer. Whether lean mass is reported as kilograms or as a share of weight lost changes which direction the result points. None of these is cheating. They are judgement calls, and they are the reason the most careful of the four graded its own certainty as low to very low and declined to conclude anything.
What none of them disagree about is the direction training pushes. Where a resistance training arm existed, it produced the lowest share of weight lost as lean tissue in the analysis. That arm was almost always measured without anybody taking a GLP-1, so applying it to a person on one is an extrapolation, and we are telling you it is an extrapolation rather than dressing it as a finding.
The lesson from the drug that worked and did nothing
Bimagrumab is the most instructive result on this page. In 180 older adults with sarcopenia it did exactly what it was designed to do, adding six percentage points of lean body mass over placebo with a p value below 0.001. Then every functional endpoint came back null. Walking speed, six minute walk distance and the physical performance battery all failed to separate from placebo.
Both groups did improve, and the authors credit the diet and the light home exercise that everybody in the trial received. Which is the uncomfortable summary: in that trial the nutrition and the movement did the work, and the drug that reliably built tissue did not add to it.
Carry that across to the question people are actually asking about GLP-1s. If your goal is a better number on a body composition scan, that is achievable and it has been achieved pharmacologically. If your goal is to be able to get off the floor at eighty, the scan is a proxy for that and this trial is a warning about how loose a proxy it can be.
When the answer arrives
More than a dozen trials are registered that combine resistance training, protein targets or both with these drugs, and none has posted results. The best designed of them randomises people on semaglutide or tirzepatide to resistance training, to a higher protein target, to both, or to neither, with thigh muscle measured by MRI as the primary outcome, and it is run by a university group rather than a manufacturer. Its primary completion date is August 2029. Until then, anyone stating confidently what training does on these drugs is reasoning from adjacent evidence, which is reasonable, and is not the same as knowing.
Questions people ask about this
- Is there research on GLP-1 drugs and training?
- A great deal, and it is growing fast. There are body composition substudies from the major drug trials, four systematic reviews published in 2026 alone, randomised trials of drugs designed to preserve muscle during weight loss, and a landmark randomised trial combining exercise with liraglutide. What does not exist is a published randomised trial of progressive resistance training during semaglutide, tirzepatide or retatrutide. Those two statements are both true and they are usually confused for each other.
- Does tirzepatide have exercise data?
- One randomised trial combined tirzepatide with resistance and aerobic training. It was retracted in 2025 and PubMed flags the original as a retracted publication, so its results cannot be used. Its numbers still circulate. Tirzepatide does have body composition data from SURMOUNT-1 by DXA and SURPASS-3 by MRI, neither of which involved a training intervention.
- What about retatrutide?
- Its phase 2 body composition substudy in type 2 diabetes measured fat and lean mass by DXA in the subset with paired scans. There is no exercise intervention in it, and no randomised training data for retatrutide exists at all.
- So should I lift while taking one of these?
- The reasoning for it is strong and the direct trial evidence is absent, and it is worth knowing which is which. Resistance training has randomised evidence that it builds and preserves muscle in people who are not on these drugs, including in an energy deficit. Nobody has yet published a trial putting that together with a GLP-1. The first one designed to answer it properly reports in 2029.
- If a drug can add lean mass, is that not simpler than training?
- One of the most useful results here says no. Bimagrumab added 6 percentage points of lean body mass over placebo in older adults with sarcopenia, and improved not one measure of physical function: walking speed, six minute walk and the physical performance battery were all null. Adding tissue to a scan and improving what a person can do turned out to be different things.