This is the safety question with the widest gap between what a reader has heard and what has been measured. The reports came first, the regulatory investigations followed, and the large studies arrived afterwards and mostly disagreed with the premise. That sequence is worth understanding before reading any single number below — the same three stages ran for sudden vision loss, and knowing which stage a question has reached tells you how much weight its headline deserves.
The cohort that answered the regulators
A study of electronic health records assembled 240,618 patients with overweight or obesity who were prescribed either semaglutide or a non-GLP-1 anti-obesity medication, matched them on propensity score, and followed them for six months [1]. Incident suicidal ideation carried a hazard ratio of 0.27 on semaglutide, and recurrent ideation 0.44 (95% CI 0.32 to 0.60). The finding held across sex, age and ethnicity, and replicated in a separate population of 1,589,855 patients with type 2 diabetes.
A hazard ratio below 1.0 means the event happened less often in the semaglutide group. The authors’ own conclusion is phrased negatively — the findings do not support a higher risk — which is the correct strength of claim for this design.
Semaglutide against the rest of its own class
A 2026 analysis of a global records network covering more than 192 million patients ran two matched comparisons [2]. Semaglutide against earlier GLP-1 receptor agonists, across 80,115 matched pairs, showed lower first-year rates of depression (HR 0.811, 95% CI 0.770 to 0.855), anxiety (HR 0.915, 95% CI 0.871 to 0.961) and suicidal ideation (HR 0.488, 95% CI 0.339 to 0.702).
Tirzepatide against semaglutide, across 85,546 matched pairs, showed no meaningful difference on the composite psychiatric outcome in either year one (HR 0.984, 95% CI 0.950 to 1.019) or year two (HR 1.002, 95% CI 0.960 to 1.046). A nominally higher anxiety hazard appeared for tirzepatide in year two (HR 1.052, 95% CI 1.001 to 1.106), and the authors themselves flag it as fragile given the number of comparisons made.
People who were already at risk
A 2026 target-trial emulation in 606,434 US veterans with type 2 diabetes looked specifically at patients who already had a substance use disorder [3]. Among them, starting a GLP-1 receptor agonist rather than an SGLT-2 inhibitor was associated with a lower rate of suicidal ideation or attempt (HR 0.75, 95% CI 0.67 to 0.83), a difference of roughly 10 fewer events per 1,000 people over three years.
What to do with this
None of these studies randomized anybody to semaglutide to find out. People who get prescribed one drug differ from people who get prescribed another, and propensity matching reduces that problem without eliminating it. The honest summary is that the signal which prompted the investigations has not reproduced in large matched populations, and that this is not the same as proof it does not exist for any individual.
Practically, this is a reason to want a prescriber you can reach. Most of the sellers tracked here publish nothing about who reviews a case or how often, which is the silence counted in what sellers will not tell you and enumerated in the questions they leave open. Anyone with a psychiatric history should be raising it with a clinician who can act on it rather than an intake form, and anyone deciding whether to stop should read what happens after stopping first. The related question of whether these drugs affect drinking is covered in semaglutide and alcohol use.