<p>Genome-wide association studies have identified thousands of single-nucleotide variants that are associated with complex traits, including cardiometabolic diseases, cancers and neurological disorders. Polygenic risk scores (PRSs), which aggregate the effects of these variants, can help to identify individuals who are at increased risk of developing such diseases. As PRSs are typically only weakly associated with conventional risk factors for these diseases, they have incremental predictive value and are beginning to be incorporated into clinical practice to guide early detection and preventive strategies. However, challenges to their use — such as suboptimal precision, poor transferability across diverse populations and low familiarity among patients and providers with the concept of polygenic risk — must be addressed before their broader clinical adoption. This Review explores the current state of the field, highlights key challenges and outlines future directions for the use of PRSs to improve risk prediction and to advance personalized prevention in clinical care.</p>