Two bodies of published work describe what these drugs do to disordered eating, and they disagree. One is a set of small treatment studies in which binge-eating scores fall. The other is a database of spontaneous adverse-event reports in which binge eating, fear of eating and self-induced vomiting all come back above their expected rates for the same drug class. Neither can settle the other, because they are not measuring the same thing with the same instrument.
No GLP-1 receptor agonist is approved anywhere for an eating disorder, and the pivotal weight-loss trials were not designed to answer this question. The evidence that exists is thin in a specific way worth naming before any of it is quoted, and it is thinner still for anorexia nervosa than for binge eating. What a cash-pay purchase does not come with is set out in the telehealth red flags article.
The treatment literature is twelve small studies
A 2026 systematic review registered with PROSPERO screened 1,125 records and included 12 studies of GLP-1 receptor agonists or dual GIP/GLP-1 agonists in people with diagnosed binge-eating disorder or binge-eating behaviors. Sample sizes were generally under 75 participants. Across heterogeneous designs the review found consistent reductions in Binge Eating Scale scores, in binge frequency and in reported remission rates, alongside weight reductions spanning −3 to −24 kg, with adverse effects that were primarily gastrointestinal.[1]
A 2025 meta-analysis pooled the quantitative subset: five studies and 182 participants. Weight fell 3.81 kg more than control (95% CI −5.14 to −2.49, I² 59.88%), body-mass index fell 1.48 kg/m², waist circumference fell 3.14 cm, and the Binge Eating Scale improved by 8.14 points (95% CI −13.13 to −3.15). Its authors close by noting that data on eating disorders other than binge-eating disorder, and on the long-term effects of these drugs, are what the field does not have.[2]
The individual studies are the size that implies. A 2015 pilot randomized 44 non-diabetic adults with obesity and subclinical binge eating to liraglutide or control for 12 weeks; binge-eating score improved alongside weight, and plasma ghrelin rose, which the authors flag as something that may diminish the weight effect once the drug stops.[3] A 2020 open-label pilot assigned adults with type 2 diabetes and binge-eating disorder to dulaglutide 1.5 mg or gliclazide 60 mg for 12 weeks, and the dulaglutide arm showed a greater fall in binge-eating behavior (P < 0.0001).[4] A 2023 retrospective cohort compared semaglutide against lisdexamfetamine or topiramate and found larger Binge Eating Scale reductions in the semaglutide-only group.[5]
Twelve weeks is not a course of treatment
The durations above are 12 weeks, 12 weeks and a retrospective chart window. None of these studies followed anyone after the drug was withdrawn, which means the literature contains no measurement of what happens to binge-eating symptoms when treatment ends. The weight side of that question has been answered in randomized withdrawal trials — see what the withdrawal trials measured — and the eating-behavior side has not been asked at all.
The registered pipeline is correspondingly small. A search of ClinicalTrials.gov on September 15, 2026 for trials naming binge-eating disorder together with semaglutide, liraglutide, tirzepatide, dulaglutide or exenatide returned five records. One is terminated: a phase 3 liraglutide study that enrolled 36 participants and stopped because it was not meeting recruitment goals. The first head-to-head against the only drug the FDA has approved for binge-eating disorder is a phase 2 trial of tirzepatide against placebo or lisdexamfetamine, with 105 participants estimated and primary completion in December 2027.
The same behavior, measured the other way
A 2024 pharmacovigilance study pulled 181,238 adverse-event reports naming GLP-1 receptor agonists from the FDA Adverse Event Reporting System between 2004 and the first quarter of 2023, of which 8,240 were psychiatric. Disproportionality analysis flagged eight categories. Four of them concern eating: eating disorder at a reporting odds ratio of 1.57 (95% CI 1.40 to 1.77), binge eating at 2.70 (95% CI 1.75 to 4.16), fear of eating at 3.35 (95% CI 1.65 to 6.78) and self-induced vomiting at 3.77 (95% CI 1.77 to 8.03). Median time to onset across all psychiatric events was 31 days.[6]
A reporting odds ratio is not an incidence and not a risk. It says a term appeared in this database more often, relative to everything else reported for these drugs, than it appeared for other drugs — and spontaneous reporting is shaped by publicity, by litigation and by who bothers to file. It cannot establish cause. What it does register is that binge eating and self-induced vomiting were being written down on these drugs at a rate the database noticed, over the same years the treatment studies were reporting the opposite direction. Both are findings. Neither is an answer.
The comparative cohort evidence is limited and does not resolve it either. A 2026 retrospective cohort of adults with class 3 obesity in a publicly funded Sydney program found no significant difference in Eating Disorder Examination-Questionnaire Short scores between the 59 of 666 participants already on a GLP-1 and those who were not; among 31 who started one, weight fell from a median 131.0 kg to 120.0 kg over 12 months with no statistically significant change in EDE-QS, K10 or DASS-21 scores, and nobody stopped for eating-disorder symptoms.[7] Thirty-one people is the size of that reassurance.
A larger comparison exists in adolescents. A 2026 electronic-health-record cohort propensity-matched 9,222 adolescents in each arm, comparing new users of semaglutide, liraglutide or tirzepatide against lifestyle intervention over 30 to 1,095 days, and reported no significant difference in eating disorders.[8] Three caveats travel with that. Eating disorders were one of six secondary outcomes; the outcome is a diagnosis code rather than a screening instrument, so it counts recognition rather than symptoms; and the paper publishes no hazard ratio for it. The sleep half of the same study is covered in the article on sleep, and the adolescent approval picture in the adolescent evidence article.
The favorable numbers came from trials built to exclude the question
The public ClinicalTrials.gov record for SURMOUNT-1 lists its exclusions verbatim, and two are psychiatric: “History of significant active or unstable major depressive disorder (MDD) or other severe psychiatric disorder within the last 2 years” and “Any lifetime history of a suicide attempt.” The registry record for STEP 1 is shorter and names no psychiatric criterion at all for its main phase, listing only an HbA1c ceiling and a recent-weight-change limit; the full protocol is not in that record, so its silence is not evidence either way. In neither record does the string “eating disorder” appear.
What those trials did instead is visible in the post hoc analyses. A 2026 pooled analysis of SURMOUNT-1, SURMOUNT-2 and SURMOUNT-3 covering 4,056 participants reports its population in its own title: people with obesity and no known major psychopathology. Depressive symptoms were tracked with the PHQ-9 at a mean baseline of 2.7, and suicidal ideation and behavior with the Columbia-Suicide Severity Rating Scale at every visit. Its conclusion asks for further study of safety in people with significant psychiatric illness.[9] That is a structured screening and monitoring apparatus running behind every efficacy figure this market quotes.
Whether any part of it is reproduced at a given cash-pay checkout is a per-seller question this site answers in the provider write-ups rather than a generalization. What can be said is that the compounded semaglutide and tirzepatide dispensed across this market hold no FDA approval, and the agency does not review a compounded preparation for safety, efficacy or quality before it reaches a patient — the contents question is in what a compounded vial contains. Who a prescriber should decline outright is covered in the contraindications article.
Anorexia nervosa is not a thin evidence base; it is an empty one
A ClinicalTrials.gov search on September 15, 2026 for interventional studies of any GLP-1 or dual agonist in anorexia nervosa returned zero records. The same endpoint returns nine for anorexia nervosa with olanzapine and 334 for obesity with semaglutide, so the zero reflects the field rather than the query.
What exists instead is case reports, which describe one person each and establish that something happened, not how often. A 2026 report describes a 21-year-old woman with two months of dieting and intermittent binge-purge symptoms who obtained tirzepatide through an online medical service, told the consultation she was overweight, and self-administered 2.5 mg weekly for four weeks then 5.0 mg weekly for a month. Her weight fell from 47 kg to 41 kg, a body-mass index of 17.9 to 15.6 kg/m². She was admitted with nausea, bilious vomiting and presyncope, with 3-hydroxybutyrate at 2,057 µmol/L and glucose at 79 mg/dL, and was diagnosed with starvation ketosis and then with anorexia nervosa, binge-eating/purging type. At later follow-up her ketones had normalized and her drive for thinness had not.[10] A 2025 report describes an adolescent girl admitted with bradycardia and pericardial effusion who disclosed semaglutide use during the stay, kept losing weight after stopping it at three months, and was later admitted to an eating-disorder ward.[11]
Bulimia nervosa appears once in the posted SURMOUNT-1 results as a serious adverse event: one participant in the 10 mg arm of 636 at risk, none of 643 on placebo and none in the other two tirzepatide arms.[12] A single event across 2,539 randomized participants supports no rate and no comparison. It is the only eating-disorder event the pivotal tirzepatide trial reported at all.
What the labels say, which is nothing
The Wegovy and Zepbound prescribing information were read from DailyMed on September 15, 2026. Across 238,724 and 192,410 extracted characters, “eating disorder,” “anorexia,” “bulimia” and “binge” each appear zero times in both. The same extraction returns “thyroid C-cell” 21 and 19 times and “Acute Pancreatitis” 12 times in each, so the zeros are not a broken text layer.
Both labels also record, in their Recent Major Changes tables, that a Warnings and Precautions section titled Suicidal Behavior and Ideation was removed in February 2026. The studies behind that regulatory history are in the mental health article, and the reward-pathway work that makes this class interesting to psychiatry at all is in the alcohol use disorder article. Neither label has ever carried eating-disorder language in either direction — not as a warning, and not as a contraindication a prescriber could point to.
What the eating-disorder field is actually asking for
A 2026 rapid review searched to January 2025 for studies reporting eating behaviors or eating-disorder risk in people treated with these drugs for obesity or type 2 diabetes. Of 1,597 records screened it found 25 studies, two in adolescents (n = 275) and 23 in adults (n = 8,722). Two reported eating-disorder adverse events and two reported none. For most measures, eating behaviors improved or stayed the same, and the review’s stated conclusion is that comprehensive assessment of eating behaviors is what is needed to understand the benefits and the risks.[13]
A 2026 spotlight article in a specialist eating-disorders journal frames the mechanism problem directly: the same appetite suppression that reduces binge eating and food preoccupation in some people may reinforce restriction, avoidance of regular eating and compulsive weight control in others. Its six proposed priorities begin with routine eating-disorder screening before and during treatment and include risk stratification, validated monitoring tools, and discontinuation planning — a list that describes what does not yet exist.[14]
That is the honest state of it. Binge eating falls in small, short, mostly unblinded studies, and nobody has measured what happens after the drug stops. Spontaneous reports carry eating-disorder terms above their expected rates, and cannot establish cause. The one large comparative cohort found no difference on a diagnosis code in adolescents. Anorexia nervosa has been studied in no registered trial. A history of any of these conditions is a fact a prescriber needs before the first dose, and it is not something a page or a checkout can weigh.