AI-powered analysis of 772 patient reviews found semaglutide users taking the drug for over 60 days lost an average of 32.2 pounds, with nausea (47%), headache (18%), and vomiting (14%) as the most common side effects.
32.2 lbs average weight lossMean weight loss in semaglutide users reporting treatment duration over 60 days, extracted from AI analysis of 772 patient reviews
What the researchers found
Among 95 reviewers reporting both weight loss amounts and treatment duration, users taking semaglutide for more than 60 days achieved mean weight loss of 32.2 ± 3.1 lbs (14.6 kg).
The most frequently mentioned side effects were: nausea (46.9%), headache (18.4%), vomiting (14.3%), fatigue (9.2%), and dizziness (4.8%). Sentiment analysis showed the highest satisfaction scores in the ≤30-day group (mean: 3.38 on a 5-point scale). Topic modeling identified dominant themes: appetite suppression, medication cost and access barriers, and long-term experiences. Cluster analysis revealed distinct user profiles including a 'super-responder' group and a side-effect-burdened group.
Why it matters
Clinical trials are conducted under ideal conditions with selected patients. Real-world patient experiences can differ significantly. This AI-powered approach to mining patient reviews provides a scalable way to understand how peptide drugs like semaglutide actually perform in diverse populations, including patient satisfaction, common complaints, and access challenges that clinical trials don't capture. The identification of super-responder and side-effect-burdened clusters could help personalize treatment expectations.
How the study worked
Retrospective cross-sectional analysis of 772 user-generated reviews of semaglutide from Drugs.com (July 2021 - March 2025). Sentiment analysis used a BERT transformer model on a five-point scale. Topic modeling employed Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA) to identify dominant themes. Cluster analysis segmented users by weight loss outcomes and side effect severity. A subset of 95 reviewers with explicit weight and duration data was analyzed for efficacy metrics.
What this study cannot tell us
Patient reviews on Drugs.com are self-selected and not representative of all semaglutide users. Only 95 of 772 reviewers reported specific weight loss and duration data. Reviews may be biased toward extreme experiences (very positive or very negative). No verification of reported weights, doses, or treatment durations was possible. The BERT sentiment model may not perfectly capture nuanced medical experiences. Side effect frequencies from reviews are not comparable to clinical trial incidence rates due to different denominators and reporting methods.
How to read the evidence
This is a novel methodological approach using AI to analyze unstructured patient data. While it provides valuable real-world insights, the self-selected nature of online reviews, lack of verification, and small efficacy subset (n=95) mean results should be considered hypothesis-generating rather than definitive.
When this study was published
Published in 2025 analyzing reviews through March 2025, this study captures recent real-world patient experiences during the peak adoption period of semaglutide for weight management.
The bigger picture
This study demonstrates how AI and NLP can transform unstructured patient narratives into actionable pharmacovigilance data. As GLP-1 peptide drugs are prescribed to millions, traditional clinical trial monitoring can't capture the full range of patient experiences. AI analysis of online reviews, social media, and patient forums could become a standard tool for post-marketing surveillance of peptide therapeutics, complementing formal adverse event reporting systems.
Questions still open
- Could AI analysis of patient reviews serve as an early warning system for rare or unexpected side effects of peptide drugs?
- What distinguishes 'super-responders' from side-effect-burdened users at a clinical or genetic level?
- How do AI-derived real-world efficacy estimates compare to formal observational studies and registries?
Common questions
How much weight do real people lose on semaglutide?
What are the most common side effects people report with semaglutide?
Read the original research
Artificial intelligence and natural language processing of patient narratives to evaluate semaglutide for weight loss.
Annals of epidemiology, 111, 9-13
Citation
Bhagavathula, Akshaya Srikanth. (2025). Artificial intelligence and natural language processing of patient narratives to evaluate semaglutide for weight loss.. Annals of epidemiology, 111, 9-13. https://doi.org/10.1016/j.annepidem.2025.09.003