The number everyone quotes for semaglutide is around fifteen per cent of body weight. The number a 2025 Cochrane review actually reports, pooling fifteen randomized trials in 8,651 adults, is a mean difference of −10.73 per cent against placebo, with a 95 per cent confidence interval from −12.24 to −9.21 [1]. That is a large effect, it carries high-certainty evidence, and it is smaller than the figure a seller’s landing page will show you. Before comparing what any of them charge for it, it is worth knowing what the drug was measured to do.
What is certain, and what is not
The review graded each outcome separately, and the grades fall away sharply as the outcome moves from the scale to the person. Percentage body weight is high certainty. So is the proportion reaching five per cent weight loss, a risk ratio of 2.68 (95% CI 2.30 to 3.12) across twelve studies and 7,458 participants. Quality of life is moderate certainty, and the measured difference on the SF-36 physical functioning score is 2.12 points (95% CI 0.95 to 3.28) — statistically present and small enough that a reader should ask what two points feel like. Serious adverse events are graded very low certainty, with a risk ratio of 1.01 and an interval from 0.78 to 1.29 that includes no difference in either direction.
This is the shape of the evidence base, and it is not the shape of the marketing. The thing measured best is the number on the scale.
The long-term figures are thinner than they look
At 26 months the review reports a mean difference of −11.11 per cent (95% CI −16.47 to −5.75), which sounds like confirmation until you read what sits beside it: two studies, and an I² of 94 per cent. That statistic describes how much of the variation between the two results is more than chance, and ninety-four per cent is close to the ceiling. Two trials that disagree that strongly are not a settled long-term answer, and the review grades this outcome moderate rather than high for exactly that reason.
What happens after the drug stops is a different question again, and one the trials mostly do not follow. We have written separately about what the regain data shows.
Who paid for the evidence
Seventeen of the eighteen included trials reported that the drug’s manufacturer had a major role in their design, conduct, analysis or writing. The review names this explicitly in its conclusions as raising important concerns about conflicts of interest, and it identifies 46 ongoing studies that may change the picture.
This is not a reason to disregard the findings. Industry-run trials are how most drugs are tested, and a high-certainty grade from Cochrane already accounts for a great deal. It is a reason to notice that the independent replication most readers assume exists behind a number this famous has, so far, largely not happened.
A note on the cardiovascular figures
The review reports major adverse cardiovascular events at a risk ratio of 0.63 (95% CI 0.44 to 0.90) at medium-term follow-up and still describes the effect as possibly little to no absolute difference. That reads like a contradiction and is not one. The ratio is relative; the absolute number of events in a mostly healthy trial population is small, and the certainty grade is low. A thirty-seven per cent relative reduction in a rare outcome can be a handful of events either way.
What to do with this
Take the ten to eleven per cent figure as the honest central estimate for what the trials measured, expect less certainty the further an outcome sits from body weight, and treat the long-term numbers as provisional. Then look at what a seller will actually tell you about the product before you buy it — our census of what goes unpublished and the questions left open are the practical half of this.