Fifty people with moderate to severe alcohol use disorder took a daily pill for eight weeks. Half got oral semaglutide, the pill form of Novo Nordisk's weight-loss and diabetes drug, and half got a placebo. The one measure the trial was built around, a laboratory test of alcohol craving, did not separate the two groups. Almost everything else did.
The phase 2 trial, published July 29 in the American Journal of Psychiatry ↗, is one of the first randomized, placebo-controlled tests of semaglutide ↗ in people who were actively seeking help for a drinking problem. That last detail matters. Earlier signs that GLP-1 drugs curb alcohol use came mostly from animals, from health-record studies, and from a trial in people who were not trying to quit and drank less heavily. This one enrolled a harder group and used the swallowed form of the drug, sold as Rybelsus, rather than the injection. The dose stepped from 3 mg a day for four weeks to 7 mg a day for four more.
The endpoint it was built to hit is the one it missed
The pre-registered primary outcome was cue-elicited craving, measured at week 6 by putting alcohol in front of participants in a lab and asking how badly they wanted a drink. Semaglutide did not beat placebo on that test. It also did not significantly lower the average number of drinks per day, one of the two headline secondary measures the researchers had named in advance.
So on the outcomes the team flagged up front as the ones that would define success, the drug came up short. A small, short trial with a null primary endpoint is the kind of result that usually cools a program, or ends it.
What moved instead
Then there is the rest of the data. Semaglutide significantly cut the number of heavy drinking days over the last four weeks of treatment (the modeled reduction had a 95 percent confidence interval of -1.01 to -0.15, meaning the effect stayed on the drinking-less side of zero). It cut how much people drank on the days they did drink. It lowered craving as people reported it in ordinary life, away from the lab. It reduced the count of alcohol-related problems, the arguments, injuries, and regretted decisions a standard questionnaire tallies. More people on the drug dropped at least one rung on the World Health Organization's risk-drinking ladder than on placebo.
One unplanned finding stood out. Semaglutide also cut the number of days people used cannabis. Nobody dosed these patients to change a cannabis habit, and the trial cannot say why it happened, but it fits a broader suspicion that GLP-1 drugs reach into the brain's shared reward machinery rather than any single appetite.
How to read a split result
The clean way to read this is uncomfortable for both camps. The measure built to prove the mechanism, craving on demand in a lab, is the one that failed, which should temper anyone selling semaglutide as a craving eraser. The measures that map onto a person's actual week, fewer heavy days and fewer bad outcomes, are the ones that moved, which is what a patient would notice.
Fifty people over eight weeks is a small, short study, and secondary endpoints that turn up positive when the primary fails have to be treated as leads, not proof. The more measures a trial tests, the more likely one drifts positive by chance. The authors' own framing is modest: the result confirms an earlier signal in lighter, non-treatment-seeking drinkers and, in their words, means "continued development of semaglutide for AUD is warranted."
That is the honest headline. Not that a weight-loss pill fixes alcohol use disorder, but that in the population most trials avoid, it moved enough of the right needles to justify a larger study. Semaglutide ↗ works through the GLP-1 receptor ↗, the same switch behind its effects on appetite and blood sugar. Whether that receptor is also a lever on drinking is now a question worth the bigger trial.