Semaglutide Changed Decision-Making in a Depression Trial. That Does Not Mean It Treats Depression.

The headline is intriguing. The endpoint is not a depression score.

A new randomized-trial analysis found that oral semaglutide changed how a small group of adults with major depressive disorder made choices in a computer task. Participants showed more exploitation of options expected to pay off, more directed exploration of uncertain options, and more tendency to repeat a previous choice. Those are measurable shifts in decision behavior—not evidence that semaglutide treats depression, lifts mood, or improves day-to-day functioning.

That distinction matters because “brain effects” can sound like a treatment promise. The paper is interesting precisely because it asks a narrower question: can a metabolic drug alter a specific kind of reward-learning behavior in people already receiving care for major depressive disorder? It does not answer whether patients feel less depressed.

What the researchers tested

Zhang and colleagues analyzed data from a randomized clinical trial involving 72 adults diagnosed with major depressive disorder. Thirty-five were assigned oral semaglutide and 37 placebo. Participants completed a four-armed “restless bandit” task at baseline and after 16 weeks. In this task, the value of each option changes over time; participants must decide whether to keep choosing a currently rewarding option or explore alternatives that might be better. The research team used a trial-by-trial reinforcement-learning model to estimate those strategies.

This was a secondary analysis. The parent trial examined semaglutide in people with major depressive disorder and cognitive dysfunction; this report looked at a particular computational task within that trial. It was not a separate, independently recruited replication, and the bandit-task outcomes should not be confused with a clinical depression scale.

Three modeled behaviors shifted

Compared with placebo, semaglutide was associated with increased exploitation—choosing the option with the highest expected reward (average treatment effect 0.0242; p<0.001). Directed exploration, meaning a greater tendency to try an option with higher uncertainty, also increased (ATE 0.0619; p=0.013). So did perseveration, a value-independent tendency to repeat the previous action (ATE 0.3017; p=0.013).

At first glance, exploration and exploitation may seem contradictory. They are not necessarily opposites in a person’s overall behavior: the model estimates several choice tendencies, and both can change across different situations. Nor is “more” automatically better. Exploration may help someone learn; exploitation may help use known information; perseveration can mean consistency or rigidity depending on the setting. The task does not tell us which pattern would be beneficial in real life.

The authors reported that the modeled effects were moderated by baseline characteristics but remained independent of weight loss during the trial. That is a useful clue for future research, not proof of a direct brain mechanism. A statistical association that is not explained by measured weight change does not establish which biological pathway caused it, or rule out other pathways and limitations in the analysis.

Why this is not an antidepressant result

The study did not establish a reduction in depressive symptoms, remission, prevention of relapse, or improved quality of life from this decision-task analysis. It did not show that semaglutide can replace antidepressants, psychotherapy, or other evidence-based care. The authors’ conclusion is appropriately bounded: semaglutide modulated explore/exploit behavior in this sample, expanding questions about metabolic processes and decision-making in depression.

A companion secondary analysis from the same Toronto research program, published in JAMA Psychiatry, examined effort-based decisions in the same 72-person trial. It reported changes in willingness to exert effort for rewards. Taken together, the reports suggest a research signal around reward-related behavior. But because they draw on the same small parent trial, they are related analyses—not two independent trials confirming a clinical mental-health benefit.

The study population also matters. Participants had major depressive disorder and were recruited through a university mood-disorders program; the parent trial focused on people with overweight and cognitive dysfunction. Results from this selected group cannot automatically be generalized to all people with depression, people without overweight, or people taking semaglutide for other reasons. The 16-week follow-up does not tell us whether the task differences last, matter outside a laboratory, or predict meaningful recovery.

There is another interpretive limit: computer-task performance is not a direct readout of “motivation” in the full human sense. A participant may make more exploratory choices in a structured task without becoming more curious, energetic, or socially engaged in daily life. The model translates choices into parameters; those parameters are useful for testing theories, but their clinical meaning needs validation against outcomes patients notice and care about.

The report also does not establish that every participant changed in the same way, or that any change was large enough to matter outside the task. Group-level estimates can conceal varied individual responses. Before turning a modeled effect into a care recommendation, researchers need to show a repeatable link between the task measure and a benefit such as better functioning, fewer depressive symptoms, or a more durable recovery.

The wider GLP-1 and mental-health question

GLP-1 medicines interact with systems involved in appetite, reward, and metabolism, so researchers have good reason to examine behavior beyond body weight and glucose. But plausible biology is not clinical efficacy. Reward tasks can help test mechanisms; they cannot stand in for patient-centered outcomes such as sustained symptom improvement, functioning, safety, and ability to stay on treatment.

Oria has previously reviewed broader evidence on GLP-1 medicines and anxiety or depression. This new paper adds a specific experimental angle—modeled decision behavior—rather than resolving that broader question. Readers can start with the existing overview at https://oriabiostack.com/blog/glp1-anxiety-depression-mental-health-data-2026. The sensible takeaway is neither “these drugs are antidepressants” nor “the finding means nothing.” It is a small, hypothesis-generating signal that deserves a properly powered clinical test with prespecified mood and functioning outcomes.

What a decisive next study needs

A follow-up should recruit a larger and more diverse group, prespecify the decision-task outcomes, and measure depressive symptoms and everyday functioning alongside weight and metabolic changes. Researchers should report adverse events and treatment discontinuations clearly, test whether behavioral changes predict meaningful clinical outcomes, and distinguish medication effects from weight-loss-mediated effects without overstating what statistical models can prove. Replication by an independent team would help establish whether this is robust rather than a feature of one dataset or analytic approach.

Until then, anyone taking semaglutide should use it for an appropriate, clinician-supervised indication—not as a self-directed treatment for depression or motivation. People with depression should not stop or change psychiatric treatment based on a computational-task result. If mood symptoms are worsening, speak with a qualified clinician; urgent or suicidal thoughts warrant immediate local crisis or emergency support.

Evidence grade

Grade C — Analyzed randomized-trial data provide a controlled but small-sample signal on modeled decision behavior. This is a secondary analysis of 72 participants, uses task-based outcomes rather than clinical depression endpoints, and has not established symptom relief or real-world benefit. The related effort-based report comes from the same parent trial, so it is not independent replication.

Sources

Zhang Z, et al. “Semaglutide modulates explore/exploit decision-making in major depressive disorder: A randomized clinical trial.” Psychological Medicine. 2026. PMID: 42904347. https://pubmed.ncbi.nlm.nih.gov/42904347/ | DOI: https://doi.org/10.1017/S0033291726106412

Gill H, et al. “Semaglutide and Effort-Based Decision-Making in Major Depressive Disorder: A Randomized Clinical Trial.” JAMA Psychiatry. 2026;83(7):751–754. https://pubmed.ncbi.nlm.nih.gov/42054055/ | https://jamanetwork.com/journals/jamapsychiatry/fullarticle/2848047

ClinicalTrials.gov. NCT04466345, Semaglutide for the Treatment of Cognitive Dysfunction in Major Depressive Disorder. https://clinicaltrials.gov/study/NCT04466345

Medical disclaimer

This article is for educational purposes only and is not medical advice, diagnosis, or treatment. Semaglutide should be used only under the supervision of a licensed healthcare professional for an approved or otherwise legally appropriate indication. Do not start, stop, or change medication based on this article. Oria provides research information and clinician teleconsultation; individual decisions belong with your treating clinician.