What 400,000 Reddit posts reveal about Ozempic, Wegovy and other GLP-1 drugs

Artificial intelligence is increasingly transforming healthcare, from helping radiologists identify disease to accelerating drug discovery. Now researchers are deploying AI in a different way: listening to the experiences patients share online.

A new study from the University of Pennsylvania has analysed more than 400,000 Reddit posts discussing popular GLP-1 medications, including semaglutide products such as Ozempic, Wegovy and Rybelsus, and tirzepatide medicines including Mounjaro and Zepbound. The research identified several commonly discussed symptoms that may be underrepresented in clinical trials and official regulatory documentation. Published in Nature Health, the study does not show that these medicines caused the reported symptoms. Instead, it suggests that large-scale social media analysis may help uncover patient concerns and potential safety signals that warrant further scientific investigation.

Few drug classes have captured public attention in recent years like glucagon-like peptide-1 receptor agonists, commonly known as GLP-1 drugs. Originally developed to help manage type 2 diabetes, medicines such as Wegovy and Mounjaro have become widely known for their effectiveness in supporting weight loss.

As their popularity has grown, millions of people worldwide have begun taking the medications, creating an unprecedented opportunity to observe how they affect patients outside the controlled environment of clinical trials. Traditional pharmacovigilance systems rely heavily on physician reports, regulatory databases and post-marketing surveillance. However, patients increasingly discuss their experiences in online communities long before those experiences make their way into formal reporting systems.

The Penn researchers examined 410,198 Reddit posts published between May 2019 and June 2025 by 67,008 users who self-reported taking either semaglutide or tirzepatide. Using advanced large language models and computational analysis techniques, the researchers categorised symptoms described in everyday language and mapped them to recognised medical terminology. This type of analysis would have been extremely difficult only a few years ago.

Patients rarely describe symptoms using clinical language. Someone experiencing thermoregulatory changes might simply write that they are “always cold”, while another person may complain about recurrent chills or unexplained hot flushes. AI systems can now process enormous volumes of text and identify common patterns across hundreds of thousands of conversations. The researchers describe this approach as “computational social listening”, a method designed to extract meaningful health information from large-scale online discussions.

The results contained several reassuring findings. Many of the most frequently discussed symptoms corresponded closely with side effects already recognised by regulators and physicians. Gastrointestinal complaints dominated the discussions, with nausea, vomiting, constipation and diarrhoea among the most commonly reported issues. These findings broadly mirror the safety profiles already established through clinical studies and prescribing information.

However, the study also identified symptom categories that appeared frequently enough to attract attention but are less commonly highlighted in current product labelling. Among users who reported side effects, nearly four per cent mentioned reproductive symptoms, including irregular menstrual cycles, heavy bleeding and bleeding between periods. Reports relating to body temperature regulation also emerged, including chills, feeling unusually cold, hot flushes and fever-like symptoms. Fatigue proved particularly noteworthy. While tiredness is not absent from the clinical literature, it appeared as the second most commonly reported complaint within the Reddit dataset, suggesting the symptom may have a greater impact on some patients than existing datasets indicate.

Signal does not equal proof

The researchers stress that these observations do not establish causation. A central limitation of social media research is that it captures associations rather than controlled evidence. People may experience symptoms for many reasons unrelated to the medication they are taking. Other health conditions, lifestyle factors, weight loss itself or concurrent treatments could contribute.

The authors therefore caution against interpreting online reports as proof that GLP-1 medicines directly cause menstrual changes, hot flushes or temperature disturbances. Instead, they argue that repeatedly reported experiences can help generate new research questions. Pharmacovigilance has always relied on identifying early signals before conducting rigorous scientific investigation. The value of the Reddit analysis lies not in proving risk but in identifying issues that might otherwise remain overlooked.

One reason the findings attracted scientific interest concerns the hypothalamus, a small but vital region of the brain involved in regulating hunger, body temperature and reproductive hormones. GLP-1 medicines exert some of their effects through neural pathways associated with appetite regulation. Researchers suggest this biological connection provides a plausible rationale for examining whether temperature-related symptoms and menstrual changes deserve closer scrutiny. However, they emphasise that this remains a hypothesis rather than evidence of a causal mechanism. Further clinical research would be required to determine whether the observed patterns reflect genuine physiological effects, weight-loss-related changes, reporting bias or other factors.

The study highlights a broader trend in healthcare: the use of AI to complement traditional medical research. Clinical trials remain the gold standard for evaluating safety and efficacy. Yet they necessarily involve defined patient populations and fixed study durations. Rare effects or concerns that matter most to patients may not always emerge clearly until a medicine reaches widespread use. AI-driven analysis of online communities offers a potential supplementary source of information. Rather than replacing conventional clinical research, it can provide an early warning system that helps researchers identify questions worth investigating more formally. The approach may be particularly valuable when medicines achieve rapid mainstream adoption, as has occurred with GLP-1 therapies, where millions of prescriptions have been issued within a relatively short period.