Alhena tested 15 live AI shopping agents in 2026. All 15 explained how to process a return. Only 4 actually did it. Inside the widest performance gap in agentic CX.
Alhena's 2026 stress test ran 15 live shopping agents through real storefronts as ordinary shoppers: all 15 could answer, 9 could sell, 4 could complete a return or order change, and 1 remembered the shopper on a return visit.
Most 'AI' in health and wellness retail is back-office or clinical, not a shopping agent. This dated census verifies who's actually live (Thorne's Taia, HUM's Holly, Vitamin Shoppe, Hims, Grüns), who only announced intent, and who has no public assistant at all, with the evidence for each.
Wellness AI agents convert LLM-referred traffic at 4.68%, nearly double the cross-vertical average. But that measures a traffic source, not your agent. A three-tier model separating engagement, attribution, and incrementality, plus how subscription economics change the math.
Deploying an AI agent for a supplement or wellness brand is a sequence of design decisions, not a plug-in. This playbook covers the six that matter: knowledge grounding, the FDA claims boundary, quiz-to-conversation personalization, subscription flows, human escalation, and measurement.
Wellness shoppers are already asking AI their health questions. This guide maps what's actually live in the vertical, what the conversion data shows, where the FDA/FTC boundary sits, and a maturity ladder for deploying an agent your compliance team can sign off on.