Abstract
Online AI triage tools (symptom checkers) are widely deployed at the digital front door of US health systems, but little is known about how users' pre-stated care intent aligns with the AI recommendation or how that alignment shapes downstream engagement. We conducted a prospective cohort analysis of 6772 randomly selected users completing AI self-triage on either the public website (unauthenticated) or patient portal (authenticated) of a large integrated US health system (September 2023 to May 2024). Users reported their planned site of care (pre-intent), and encounters were classified as validated (matching the AI triage recommendation) or re-directed (differing); re-directed encounters were sub-classified as escalated or de-escalated. Downstream digital engagement was captured as interaction with any call-to-action. Of 6772 users, 508 (7.5%) were unauthenticated and 6264 (92.5%) authenticated; 89% of unauthenticated self-care pre-intenders were escalated by the AI, and 42% of authenticated office-visit pre-intenders were re-directed. Call-to-action interaction was approximately twice as high when the AI recommendation matched pre-intent. Of 636 non-engagers responding to a follow-up survey (9.4% response rate), all reported plans to seek care off-platform. Authentication status and pre-intent-to-recommendation alignment are strong correlates of digital care-seeking engagement, and merit targeted user-experience design.