GLP-1 drugs and alcohol: what the trials found

The short answer

In people who are overweight or obese, yes: two 2026 randomised trials found semaglutide beat placebo on heavy drinking days. In people who are not, the evidence is thin and, in the one older trial that looked, it pointed the other way. No GLP-1 drug is approved anywhere for drinking.

GLP-1 receptor agonists are engineered copies of a gut hormone called GLP-1 (glucagon-like peptide 1), built to survive in the body far longer than the real thing. Semaglutide is sold as Ozempic for type 2 diabetes and as Wegovy for weight management. Exenatide, an older and weaker member of the same family, was sold as Byetta and Bydureon.

People taking these drugs kept saying, unprompted, that they had stopped wanting to drink. That is an anecdote, and anecdotes about appetite drugs are cheap. What makes it worth an article is that the anecdote has now been put through four randomised controlled trials in people diagnosed with alcohol use disorder, which is the study design where participants are assigned by chance to the drug or to a dummy treatment so that the two groups differ only in what they got.

They disagree, and they disagree in four different ways. One found nothing at all on its main measure [3]. One changed craving and how much people drank on a drinking day, but not how often [4]. One missed its own primary craving measure and cut heavy drinking anyway [11]. The one with the strictest entry requirement, obesity, won clearly on the measure it was built for [1]. Understanding why they disagree is more useful than any single result, because the difference between them is not really about the drug. It is about who was enrolled.

Alcohol use disorder, the clinical name for a pattern of drinking a person cannot control and that is causing harm, accounts for about 5% of deaths worldwide every year [1][3], and only three medicines are approved by the US Food and Drug Administration to treat it: disulfiram, naltrexone and acamprosate [3]. That short list is why a repurposed obesity drug gets this much attention.

What the 2026 trial actually measured

A Danish trial randomised 108 people with alcohol use disorder and obesity to weekly semaglutide or a saline placebo for 26 weeks, on top of cognitive behavioural therapy. Heavy drinking days fell 41.1 percentage points on the drug and 26.4 on placebo. The difference between the arms was 13.7 points.

The trial that produced the headlines ran at a single centre in Copenhagen between June 2023 and February 2025. It enrolled 108 treatment-seeking participants, 53 women and 55 men, all of whom had moderate to severe alcohol use disorder and, as a requirement of entry, obesity. Fifty-four were assigned to semaglutide and 54 to placebo, the placebo being a saline injection. Everyone also received standardised cognitive behavioural therapy, a structured talking treatment for drinking, and 81% of them, 88 people, finished the full 26 weeks [1].

The measure the trial was built to answer is worth slowing down on, because it is the number every news story rewrote. It is the percentage of heavy drinking days in a 30-day window, where a heavy drinking day means more than 60 grams of alcohol for a man or 48 grams for a woman in a single day, counted using a validated interview method called timeline followback in which a person reconstructs their drinking day by day against a calendar [2]. So the endpoint is not "drinks" and not "days". It is the share of a month that crossed a line.

On that measure the semaglutide group fell by 41.1 percentage points from where it started (95% confidence interval -48.7 to -33.5) and the placebo group fell by 26.4 (-34.1 to -18.6) [1]. The estimated treatment difference, which is what the drug added over and above everything the placebo group also got, was -13.7 percentage points (-22.0 to -5.4, p=0.0015) [1]. A confidence interval is the range the true value is likely to sit in; this one does not cross zero, which is what makes the result statistically significant.

That gap between 41.1 and 13.7 is the most misread thing about this trial. Both arms improved enormously, which is what usually happens when people who want to stop drinking are enrolled, seen weekly and given therapy. The drug's own contribution is the 13.7. The authors report substantial effects on multiple secondary alcohol-related and bodily outcomes as well, and describe the side effects as transient, mostly mild to moderate stomach and gut problems, more common on semaglutide than on placebo [1].

The trial that came first, and the subgroup it was built on

The same Copenhagen group tested exenatide in 127 patients in 2022 and missed its primary endpoint entirely. Its one positive drinking result was an exploratory subgroup of 30 obese patients. In the 52 patients with a BMI under 25, the drug arm did worse than placebo.

Four years before the semaglutide result, the same research group ran the same design with an older drug. The 2022 trial gave exenatide 2 mg weekly or placebo for 26 weeks, again alongside cognitive behavioural therapy, to 127 treatment-seeking patients with alcohol use disorder recruited across four alcohol outpatient clinics in Copenhagen [3]. A hundred and fifty-two people were screened to enrol those 127, and only 58 of them completed the 26 weeks, a dropout the authors themselves name as a concern for the reproducibility and reliability of their findings [3].

It did not work. Heavy drinking days and total alcohol intake fell strongly in both groups, and there was no significant difference between them [3]. That is the result the trial was designed to deliver, and it was null. Worth noting alongside it: body mass index fell by 0.95 units more in the exenatide arm than in the placebo arm, so weight did move, and the drinking endpoint still did not separate [3].

Then came the exploratory analyses, the ones researchers run after the fact to see whether some group responded differently. The prespecified ones, on baseline heavy drinking, severity and geography, found nothing. An exploratory post hoc analysis by body mass index found two things. In the 30 patients with a BMI above 30, exenatide reduced heavy drinking days by 23.6 percentage points (95% CI -44.4 to -2.7, P = 0.034) and cut total alcohol intake per 30 days by 1,205 grams (-2,206 to -204, P = 0.026) against placebo [3]. In the 52 patients with a BMI below 25, exenatide increased heavy drinking days by 27.5 percentage points (4.7 to 50.2, P = 0.024), though total intake in that subgroup did not differ [3].

Hold those two findings next to the 2026 trial and the picture changes. The 2026 trial's entry criterion, obesity, is the 2022 trial's exploratory subgroup promoted to a rule [1][2]. That is a legitimate and careful way to design a follow-up study. It also means the newer trial has, by construction, nothing to say about people with alcohol use disorder who are not obese, which is most of them. The authors of the 2022 trial offered a possible explanation for the lean result, that thinner people on these drugs may see a bigger drop in blood sugar, which has been linked to increased craving, and they flagged their own subgroup findings as preliminary [3].

The 2022 trial also did something the others have not: it followed people for six months after the drug stopped. At that visit there was no difference between the groups on anything, except that the original exenatide group scored 5.1 points higher on the AUDIT drinking questionnaire than placebo (95% CI 0.9 to 9.3, P = 0.02), higher meaning worse [3].

Where the craving and the drinking part company

A US trial of 48 adults on low-dose semaglutide cut laboratory drinking, drinks per drinking day and weekly craving, but not how many days they drank. A 2026 trial of oral semaglutide in 50 adults did the reverse: it missed its laboratory craving measure and still cut heavy drinking days.

The middle trial is the most quoted and the most often flattened. Researchers at the University of North Carolina screened 504 people and randomised 48 non-treatment-seeking adults with alcohol use disorder, 34 of them women, mean age 39.9 years [4]. Non-treatment-seeking matters: these were people with a drinking problem who were not trying to quit, which is a very different population from the Copenhagen trials [4].

The doses were low. Participants got 0.25 mg weekly for four weeks, 0.5 mg for four weeks, then 1.0 mg for one week, over nine weeks of treatment [4]. For comparison, the 2026 Danish trial titrated up to 2.4 mg [1]. So whatever this trial found, it found at a top dose of 1.0 mg a week [4], well under the 2.4 mg used both for weight management and in the Danish trial [1].

Its main measure was not weekly drinking at all. It was a laboratory alcohol self-administration task, where participants get access to alcohol under observation and the researchers measure what they actually consume rather than what they report. Semaglutide reduced grams of alcohol consumed and peak breath alcohol concentration there, with medium to large effect sizes [4].

Out in ordinary life the picture was mixed, and this is the part that gets dropped. Semaglutide did not affect average drinks per calendar day or the number of drinking days. It did significantly reduce drinks per drinking day and weekly alcohol craving, and predicted greater reductions in heavy drinking over time relative to placebo [4]. A separate signal turned up in the subgroup who smoked: cigarettes per day fell more on semaglutide than on placebo [4].

Read plainly, that trial says the drug changed how much people drank when they drank, and how much they wanted to, without changing whether they drank at all. The authors' own conclusion is that this justifies larger trials, not that the question is settled [4].

The fourth trial, published in July 2026, is the mirror image and it is the one most write-ups have not caught up with. It randomised 50 treatment-seeking adults with moderate to severe alcohol use disorder to oral semaglutide, 3 mg a day then 7 mg a day, or placebo, over eight weeks [11]. Its entry criteria required a body mass index of at least 25 [12]. Its primary outcome was laboratory cue-elicited craving at week 6, and that outcome was null, as was drinks per day [11]. What did move was heavy drinking days (b=-0.580, 95% CI -1.012 to -0.148), drinks per drinking day, craving measured in ordinary life rather than in the lab, alcohol-related consequences and cannabis use days [11]. So one trial moved the laboratory measure while leaving the plain count of drinking days untouched, and the other missed its laboratory measure while cutting heavy drinking days, using two different formulations of the same drug.

Why the population studies look far more certain

Registry and health-record studies covering millions of people show large, consistent reductions in alcohol-related harm on these drugs. A 2025 meta-analysis pooled both kinds of evidence separately and found the observational effect strong and the randomised effect not statistically significant. Randomisation is the difference, because people who stay on a weekly injection differ from people who do not.

Alongside those trials sits a much larger and much noisier body of evidence, and the gap between the two is the most honest thing in this field.

A Swedish nationwide study followed 227,866 people with alcohol use disorder from 2006 to 2023, a median of 8.8 years each, comparing periods when the same individual was taking a GLP-1 drug against periods when they were not [5]. Semaglutide, in 4,321 users, was associated with the lowest risk of hospitalisation for alcohol use disorder (adjusted hazard ratio 0.64, 95% CI 0.50 to 0.83); liraglutide, in 2,509 users, came second (0.72, 0.57 to 0.92) [5]. A hazard ratio of 0.64 means roughly a third fewer such events. For context, the approved alcohol medications in the same dataset scored 0.98 (0.96 to 1.00) [5].

A Danish register study compared 38,454 new GLP-1 users against 49,222 new users of a different diabetes drug class and found a lower rate of alcohol-related events in the first three months (hazard ratio 0.46, 0.24 to 0.86), with the risk highest in the three months before treatment started, which the authors read as a possibly transient effect around initiation [6]. A 2026 analysis of 524,817 US veterans with type 2 diabetes starting either a GLP-1 drug or an SGLT-2 inhibitor found a lower risk of newly diagnosed alcohol use disorder on the GLP-1 drugs (0.82, 0.78 to 0.85), amounting to about 5.6 fewer cases per 1,000 people over three years, with similar results for cannabis, cocaine, nicotine and opioid disorders [7].

Those numbers are enormous and they are not randomised. People who are prescribed and stay on an expensive weekly injection differ from people who are not, in ways no statistical adjustment fully removes. A systematic review published in late 2025 pooled both literatures and put the tension in one place: across three randomised trials totalling 430 people, the pooled effects on alcohol consumption, drinks per drinking day and craving were all not statistically significant, while across six observational studies totalling 2,740,207 people the hazard ratio for alcohol-related events was 0.64 (0.59 to 0.69) [8]. That review searched the literature only up to 3 May 2025, so neither 2026 trial is in it [8].

Its third pooled trial is one this article has not described, because it is not an alcohol trial. It is a prespecified secondary analysis of a smoking cessation trial: 255 people were randomised to dulaglutide (another GLP-1 drug) or placebo to help them quit, and the 151 who drank were analysed for alcohol. At week 12 the dulaglutide group drank an estimated 29% less than placebo (relative effect 0.71, 95% CI 0.52 to 0.97, P = 0.04), and 90.8% of that population was obese [14].

What the brain evidence shows, and what it does not

Rodent work found semaglutide cut binge-like drinking in mice and rats and altered signalling in the amygdala. It also cut intake of other drinks, alcoholic or not. In humans, exenatide blunted the brain response to alcohol cues without that change tracking how much anyone drank.

GLP-1 receptors exist in brain regions involved in reward, which is the foundation of the idea: if a drug dials down the pull of food, perhaps it dials down the pull of alcohol through the same circuitry.

In rodents it does. Semaglutide dose-dependently reduced binge-like alcohol drinking in mice and reduced both binge-like and dependence-induced drinking in rats, and it increased inhibitory signalling in the central amygdala and infralimbic cortex of alcohol-naive rats, suggesting more GABA release [9]. The caveat is in the same paper and is rarely repeated: a similar effect was seen on the intake of other caloric and non-caloric solutions [9]. The animals drank less of everything, which is what you would expect from a drug that suppresses consumption generally, and it makes "specifically anti-alcohol" a harder claim than it looks.

The human brain data come from the failed exenatide trial, where a subgroup had scans. Exenatide significantly reduced the brain's reactivity to alcohol cues in the ventral striatum and septal area, regions central to reward, and dopamine transporter availability was lower in the exenatide group [3]. Did the people whose cue reactivity dropped drink less? No: the researchers looked for that correlation in the whole group of 127 and in the scanned subgroups and found none [3].

A measurable change in a scan is a mechanism finding. It is not, by itself, evidence that the behaviour changed, and here it explicitly was not.

Where that leaves the question

The positive trials all enrolled people who were overweight or obese, and the one trial that looked below that line found harm rather than benefit there. No regulator has approved this use, only one trial followed anyone after the drug stopped, and that follow-up found nothing had lasted.

The defensible version of the answer has four parts. First, in people who have both alcohol use disorder and obesity, a randomised trial with therapy in both arms found semaglutide reduced heavy drinking days by 13.7 percentage points more than placebo over 26 weeks [1]. That is a real, prespecified, statistically significant result from the design built to answer the question, and it is the best evidence this idea has.

Second, that result does not transfer automatically to people who weigh less, and the evidence thins out in steps rather than all at once. One rung down, the July 2026 oral semaglutide trial required a body mass index of at least 25 [12] and did cut heavy drinking days [11], so the finding now reaches the overweight band and not just the obese one. Below a body mass index of 25 there is still no trial evidence of benefit at all: what exists is an exploratory subgroup inside a null trial, and that subgroup went the wrong way [3]. The dulaglutide analysis does not fill that hole either, because 90.8% of the people in it were obese [14]. "We do not know, and the one look we have is not reassuring" is the accurate answer there.

Third, the millions-of-people numbers are not trial numbers. When they were pooled separately, the observational studies gave a hazard ratio of 0.64 and the randomised trials gave nothing significant [8]. Where those two disagree, the randomised evidence is the one that controls for who chooses to take a drug.

Fourth, and most practically: no GLP-1 receptor agonist is approved for alcohol use disorder by any regulator. Three medicines are [3], and fewer than 2% of people with alcohol use disorder receive any medication at all [4]. Real-world surveys do show people on semaglutide and tirzepatide reporting less drinking than before they started [10], but self-report from people who chose their own medication is the weakest tier of evidence here, not the strongest.

If you want to understand why a gut hormone analogue ends up acting on reward circuitry at all, our semaglutide course walks through GLP-1 signalling from the intestine to the brainstem. The honest summary for now is that this is a promising repurposing story with one good trial in one specific population, and that is a much smaller claim than the headlines built on it.