Partly because “not losing weight” and “not responding” are different things, and partly because the spread is genuinely enormous. Across 135,349 people treated with semaglutide or tirzepatide, the share who lost more than 15% of body weight ran from 10% to 34% depending on which brand they were given [1]. A published average is a poor guide to a personal result, and how soon the result can be predicted is covered in how long a GLP-1 takes to work.
The brand spread is mostly a dose spread
Super responders, defined in that analysis as people losing more than 15%, were 34% on Zepbound, 26% on Wegovy, 24% on Mounjaro and 10% on Ozempic. Ozempic and Wegovy are the same molecule, and so are Mounjaro and Zepbound. What separates each pair is the label, the approved dose range and therefore the patient. The diabetes-labeled brands are prescribed at lower doses to people with diabetes, who consistently lose less weight on these drugs than people without it.
So 10% against 34% is not a contest between two products a buyer picks off a shelf. It is a reason to check what dose a prescription is actually written for, which is the question answered across this market in the dose most people actually reach.
The stable finding sits in the middle of the distribution. Moderate response, between 5% and 15% of body weight, ran at 40% to 42% regardless of brand, which is a narrow band across four products and 135,349 people. Roughly four in ten land there whatever they take.
Weight is one measure, and often not the one that moves
A twelve-month cohort of 166 adults with obesity and type 2 diabetes scored each patient separately on blood sugar, fat mass, inflammation and grip strength [2]. The largest group, 53.0%, improved on blood sugar without meeting the fat mass threshold. Only 29.5% did both, 5.4% moved fat mass without moving blood sugar, and 12.0% met neither.
Taken across all four measures, 47.0% counted as high responders and 15.1% responded on every dimension at once. Somebody in the 53.0% would report, accurately, that the drug was not working for weight, while their blood sugar was improving the whole time.
One limitation matters more than the rest before anyone reads that cohort as a drug comparison. It is not a GLP-1 study. It followed metformin-based regimens in which a GLP-1 was one of six add-on options across 166 completers, so the subgroup taking one is small and its size is not stated. What the study establishes is that the four dimensions come apart, not how any particular drug scores on them.
Eating pattern is the next candidate explanation
If different drugs act on different aspects of eating, then matching the drug to the pattern would explain part of the spread. A clinic scored eating behavior across five domains before and after six months of treatment and reported very large effects. Hedonic eating fell at dz 2.06 with tirzepatide, and emotional eating at dz 2.04 with naltrexone-bupropion [3].
The analytical cohort was 66 people spread across five medications, which is roughly a dozen per drug, each generating five subscale scores. Very large standardized effect sizes are what small paired samples produce, because the measure divides by a within-person variability that a dozen people estimate poorly. Everyone included had already lost at least 5% of their weight by three months, so every behavioral change recorded belongs to somebody the drug was already working for.
That does not make the hypothesis wrong, and it does mean the heatmap is not yet a map of mechanisms. The useful version of the question for a prescriber is which eating pattern describes you, rather than which number in a published grid is biggest.
What to check before concluding it is not working
Four things are worth establishing in order: the dose actually reached, how long it has been held there, what is being measured, and whether the tissue lost is the tissue intended. The last of those is the one a scale cannot see, and it is set out in what a GLP-1 does to muscle.
The fifth possibility is the most common and the least discussed. A great many people stop before the dose is high enough for the question to be answerable, for reasons counted in how long people stay on a GLP-1. A drug cannot be judged a failure at a dose and duration it never reached.