Artificial intelligence is increasingly used to identify medication nonadherence and deliver digital support, but predictive accuracy does not establish that medication-taking changes. Objective: To determine whether artificial-intelligence-enabled adherence interventions improve medication-taking, examine clinical and utilization outcomes, and evaluate heterogeneity, robustness, and certainty. We conducted a protocol-driven systematic review and random-effects meta-analysis aligned with PRISMA 2020. Comparative studies published from 2017 through July 15, 2026 were eligible when an artificial-intelligence component selected, personalized, adapted, verified, escalated, or delivered support intended to change medication-taking. Prediction-only models, static reminders, simulations, protocols, and uncontrolled studies were excluded from effect synthesis. Two reviewers independently screened records, extracted data, appraised risk of bias, and assessed certainty. Searches identified 2,587 records; 703 duplicates were removed, 1,884 records were screened, and 102 full texts were assessed. Eleven reports representing 10 studies were included. Seven studies involving 934 participants entered the primary meta-analysis. Artificial-intelligence-enabled interventions improved adherence relative to eligible comparators (Hedges g=0.41, 95% confidence interval 0.17–0.65; I²=58%). The 95% prediction interval was −0.22 to 1.04. Effects attenuated after exclusion of high-risk studies and studies with unequal adherence measurement. Clinical and utilization outcomes did not establish clear downstream benefit. Certainty was low for adherence and very low for clinical, utilization, safety, and acceptability outcomes. Artificial-intelligence-enabled interventions may improve selected adherence measures, but prediction performance should not be interpreted as intervention effectiveness. Stronger comparative studies using equivalent measurement, longer follow-up, and patient-important outcomes are required. Independent replication across medicines, populations, and care settings therefore remains necessary.