Few claims about these medicines have travelled further than the idea that they quiet the urge to drink. It has produced headlines, anecdotes and a plausible mechanism. What it has not produced, until very recently, is much randomized evidence — and the two most substantial pieces published this year point in different directions.
This page reports what those studies measured and what the labels say. It does not advise anyone about their own drinking or their own prescription; that belongs with a clinician, as our editorial standards explain.
The newest trial: a missed primary and six positive secondaries
On 2026-09-01 the American Journal of Psychiatry published a phase 2 double-blind randomized trial of oral semaglutide in 50 treatment-seeking adults with moderate to severe alcohol use disorder. Eight weeks: 3 mg a day for four weeks, then 7 mg a day for four.
Its preregistered primary outcome was laboratory-based, cue-elicited alcohol craving at week 6.
It missed. Semaglutide did not significantly reduce that craving measure against placebo, and it did not significantly reduce drinks per day.
Then the secondary outcomes, all of which favoured semaglutide:
| Outcome | Effect (b) | 95% CI |
|---|---|---|
| Heavy drinking days | −0.580 | −1.012 to −0.148 |
| Drinks per drinking day | −1.177 | −2.307 to −0.047 |
| Naturalistic craving | −2.195 | −4.174 to −0.216 |
| Alcohol-related consequences | −4.618 | −8.651 to −0.585 |
| Cannabis use days | −1.434 | −2.568 to −0.301 |
Significantly more participants on semaglutide dropped at least one WHO risk drinking level (Wald χ² = 4.01).
Both halves of that belong in any honest summary. A trial nominates a primary endpoint in advance precisely because secondary outcomes are numerous and some of them will clear significance by chance; a missed primary is not a technicality. But six secondaries moving the same way, including a real-world consequences measure, is not nothing either. The authors' own conclusion is the careful one: that continued development is warranted — not that efficacy is established.
Two limits the abstract states plainly. Fifty participants is small. Eight weeks is short for a chronic condition.
Ten weeks earlier: a meta-analysis that found nothing
In June 2026, a systematic review and meta-analysis searched to 2026-04-30 and pooled every parallel-group or crossover randomized trial of a GLP-1 agonist against placebo or control in substance use disorders. Five trials, 764 participants, follow-up from 6 to 52 weeks.
| Pooled outcome | Result |
|---|---|
| Days without alcohol consumption | mean difference −1.96 days (95% CI −17.97 to 14.05), I² = 74% |
| Cigarettes per day | −0.55 (95% CI −1.76 to 0.65), I² = 45% |
| Nicotine dependence score | +0.02 (95% CI −0.32 to 0.36), I² = 0% |
None statistically significant. Three of the five trials rated low risk of bias, two with some concerns; GRADE certainty low to moderate. The authors describe a possible signal in people with both alcohol use disorder and obesity as hypothesis-generating, and call for adequately powered, agent-specific trials before any clinical translation.
Look at that first confidence interval. It runs from about two and a half weeks of extra abstinence to about two weeks less, with 74% heterogeneity. That is not a null result in the sense of "we established there is no effect". It is a null result in the sense of "the trials so far do not constrain the answer".
The observational picture, and where it disagrees
The largest relevant dataset is not a trial. Published in July 2026, an analysis of the NIH All of Us cohort took 15,447 people with at least two recorded GLP-1 prescriptions and compared drinking scores across those with current prescriptions (3,650) and those who would receive one in the future (5,642) — a comparison group chosen to resemble the treated group on the things that lead people to be prescribed these drugs at all.
Current prescription holders scored modestly lower: incidence rate ratio 0.95 (95% CI 0.91–0.99), and 0.89 (0.85–0.93) against a propensity-score-matched comparison. People with a previous prescription showed no significant difference.
An IRR of 0.95 is a small effect, and the design cannot exclude the obvious confounders. Its authors say so and call for randomized trials.
The disagreement nobody is reporting
Here is the part worth keeping, and it comes from reading the two designs against each other rather than from either alone.
The All of Us analysis broke its drinking score into its component questions and found the association with frequency of drinking — and explicitly not with drinks per occasion or with binge drinking.
The randomized trial found no effect on drinks per day, but significant effects on drinks per drinking day and on heavy drinking days. That is a per-occasion signal.
So the observational data describes people drinking less often but not less per session, and the trial data describes something closer to the reverse. Both cannot be the whole picture of one mechanism. Anyone reasoning about how these drugs might act on drinking — whether through appetite pathways, reward pathways, or something about how a person plans a week — should notice that the two best current sources have not agreed on the basic shape of the effect, only on its direction.
What the labels actually say: nothing
The other half of what people are searching for here is not about efficacy at all. It is whether drinking is safe on one of these medicines.
We queried the FDA prescribing information for Wegovy, Ozempic, Zepbound and Mounjaro through the openFDA drug label endpoint on 2026-09-08 and searched every text field for the word "alcohol". The result across all four:
| Label | Record effective | Mentions of drinking alcohol |
|---|---|---|
| Wegovy | 2024-04-23 | None. Four hits, all "alcohol swab" in the injection instructions |
| Ozempic | 2026-01-30 | None. The word does not appear |
| Zepbound | 2026-08-28 | None. Hits are alcohol swabs and benzyl alcohol as an excipient |
| Mounjaro | 2026-07-29 | None. Same: swabs, and benzyl alcohol as an excipient |
No interaction entry, no warning, no precaution, in any of the four.
Two things follow, and only two. First, a practical one: a reader who searches an ingredient list and finds benzyl alcohol in the tirzepatide multi-dose vial and KwikPen has found a preservative, not a reason to think about drinking. Second, the important one: an absence in a label is not a safety clearance. A label records what was studied and what regulators required. That these four say nothing about drinking means the question was not addressed there — not that it has been answered. Individual risk, particularly around pancreatitis history, liver disease, or co-prescribed medicines that can cause hypoglycemia, is a conversation with a prescriber and not something a database query settles.
Where this stands
A fair summary of 2026, in three lines:
- The randomized evidence is thin — five pooled trials, 764 people — and does not yet show a significant pooled effect on alcohol outcomes.
- The newest and most directly relevant trial missed its primary endpoint and moved six secondary ones, in a sample of fifty over eight weeks.
- The largest observational dataset finds a small association, in a direction the trial data does not cleanly corroborate on the measure that matters.
That is a field with a real signal and no settled answer, which is a less satisfying story than either "GLP-1s cure drinking" or "it was hype". A PubMed search on 2026-09-08 returned 73 randomized-study records on this question; five of them were solid enough to pool. The next few adequately powered trials will decide it, and this page will be updated when they report.
For the wider pattern of large observational claims about these drugs meeting disappointing trials, our page on what the GLP-1 brain research actually found documents the same collision on cognition, where the randomized programme was considerably larger and the disagreement sharper.
Sources, with dates
- Oral Semaglutide for Alcohol Use Disorder: A Randomized Clinical Trial, American Journal of Psychiatry, 2026-09-01 — PMID 42522065, doi 10.1176/appi.ajp.20260003.
- A Systematic Review and Meta-Analysis Evaluating the Role of GLP-1 Receptor Agonists in Substance Use Disorders, Cureus, June 2026 — PMID 42524101, doi 10.7759/cureus.111641.
- Association of GLP-1 Receptor Agonist Prescriptions and Alcohol Consumption in the National Institutes of Health's All of Us Cohort, Alcohol, Clinical & Experimental Research, July 2026 — PMID 42423024, doi 10.1111/acer.70357.
- FDA prescribing information for Wegovy, Ozempic, Zepbound and Mounjaro, retrieved through the openFDA drug label endpoint on 2026-09-08.
All three studies were read as published abstracts and records on 2026-09-08; effect sizes and confidence intervals are quoted as the abstracts report them.