Artificial intelligence in obesity management: clinical evidence, translational gaps, and implementation priorities—a structured narrative review

Abstract

Obesity is a chronic, relapsing disease that requires long-term lifestyle treatment, pharmacotherapy, and metabolic and bariatric surgery. Artificial intelligence (AI) is increasingly being studied across these pathways, but its clinical readiness varies substantially. This Scale for the Assessment of Narrative Review Articles (SANRA)-informed structured narrative review searched PubMed/MEDLINE, Embase, Scopus, ScienceDirect, and Google Scholar through 1 June 2026, primarily for literature published since 2015, with citation chaining used to identify earlier landmark studies. Evidence was appraised according to study design, validation, clinical utility, workflow integration, and demonstrated patient-level benefit. Direct patient-level evidence for explicitly AI-enabled obesity interventions remains limited and heterogeneous. The strongest weight or metabolic outcome evidence largely comes from multicomponent digital, automated or hybrid-care programmes, including studies in which no AI component was independently evaluated or in which diabetes and HbA1c constituted the primary clinical context and endpoint. These findings support digital-care delivery but should not be interpreted as evidence of an AI-specific therapeutic effect in obesity. AI-assisted drug discovery remains preclinical, while natural language processing of glucagon-like peptide-1 receptor agonist narratives can support signal detection but not causal inference or quantitative safety estimation. Phenotype-based treatment and a machine-learning-assisted genetic risk score suggest potential for responder stratification, but within the eligible evidence included in this review, no externally validated, prospectively implemented AI prescribing system was found to have demonstrated improved patient outcomes. In metabolic and bariatric surgery, AI may support risk prediction, operative workflow analysis, readmission stratification, weight-trajectory modelling, and digital follow-up; most studies, however, remain retrospective or internally validated and rarely assess calibration, actionable thresholds, or prospective impact. Large language models may assist education and drafting, but current evidence does not support unsupervised treatment or procedure selection. AI should therefore augment, not replace, clinician-led multidisciplinary obesity care. Translation will require independent external validation, prospective workflow evaluation, patient-centred outcomes, safety, fairness, data governance, and cost-effectiveness.