An AI system read 410,198 Reddit posts from 67,008 people taking GLP-1 drugs and pulled out two side effects that barely register in the clinical trials: changes to menstrual cycles and trouble regulating body temperature. Neither finding proves the drugs cause those problems. Both are the kind of signal that trials, with their tight enrollment and short follow-up, are built to miss.
The analysis, published in Nature Health ↗ by a University of Pennsylvania engineering team, covered posts written between May 2019 and June 2025 by users self-reporting semaglutide ↗ (sold as Ozempic, Wegovy, and Rybelsus) or tirzepatide ↗ (Mounjaro and Zepbound). The software grouped the language people used into symptom categories and counted how often each came up. Across the whole set, 43.5 percent of users mentioned at least one side effect, meaning a little under half wrote about a problem of some kind.
What lined up and what did not
The common complaints were the ones every label already lists. Nausea led at 36.9 percent, meaning more than a third of users brought it up, followed by fatigue at 16.7 percent, vomiting at 16.3 percent, constipation at 15.3 percent, and diarrhea at 12.6 percent. That the gut dominates is no surprise. These drugs slow the stomach on purpose, and the trials that got them approved reported the same pattern.
The interesting part is what sat outside that expected list. Roughly 4 percent of users described reproductive symptoms, mostly shifts in their menstrual cycle and unusually heavy bleeding. A separate cluster described problems with temperature, chills and hot flashes, the sense of a thermostat set wrong. Four percent is a small fraction, but it is a fraction that clinical trials for these drugs did not flag as its own category, which is exactly why a crowdsourced count can be useful. People report to each other what they never think to report to a prescriber.
There is a plausible place in the body for both. The GLP-1 receptor ↗ is active in the hypothalamus, the brain region that manages hunger alongside hormone cycles and body temperature. A drug that reaches that region to blunt appetite is acting next door to the machinery that runs periods and thermoregulation. That is a reason to look, not evidence that the drug is the cause.
Why the count is a lead, not a verdict
Reddit is not a patient registry. The authors are direct about it. Its users skew younger, more male, and more American than the population actually taking these drugs, and every symptom here is self-reported by someone who chose to post. There is no untreated group to compare against, so there is no way to know how many of these menstrual changes or hot flashes would have happened anyway, in people who are often losing large amounts of weight fast, which on its own disrupts cycles and hormones. The method finds correlations in what people say. It cannot separate the drug from everything else going on.
That limit is the whole point of treating this as a starting line. The same social-media firehose cuts both ways. Earlier this year a viral fear that these drugs rot teeth failed to appear in a 226,485-patient records review ↗, a case where the crowd raised an alarm the data did not support. Here the crowd is pointing at something the formal trials were too small and too short to catalog. Both stories argue the same thing: online chatter is a decent early-warning antenna and a terrible final answer.
What would settle it is dull by comparison. Prospective studies that ask about menstrual and temperature symptoms up front, and electronic health records that actually code those complaints, would turn a 4 percent murmur into either a real adverse effect or noise. Until then the honest read is that a very large read of what patients say has handed regulators two specific questions worth chasing, in a drug class now taken by millions of adults and prescribed, increasingly, to children.