A large observational cohort study published in JAMA Network Open analyzed electronic health records from 457,950 COVID-positive patients across 58 U.S. hospitals using a precision-phenotyping AI algorithm validated for diagnosis of exclusion. It identified long COVID in 16.3% of cases—roughly 18 million Americans—double prior surveillance estimates reliant on ICD code U09.9, which captured under 7%. This Mass General Brigham-led work highlights how diagnostic coding misses the majority, with 14.5% developing new chronic conditions requiring ongoing care. Unlike randomized trials, this retrospective design draws strength from its scale but carries risks of coding biases and regional variations (13.6–22.7%). The findings connect to broader patterns of undercounted post-viral illness, echoing historical gaps in myalgic encephalomyelitis/chronic fatigue syndrome tracking where similar surveillance shortfalls delayed recognition for decades. A related 2023 CDC analysis of BRFSS data estimated 6–7% prevalence but relied on self-report without longitudinal validation, while a 2024 Lancet Infectious Diseases meta-analysis of 1.2 million patients noted persistent quarterly incidence rises, aligning with the current study's projection of continued growth absent interventions. These gaps suggest health systems must integrate AI-driven phenotyping into routine surveillance to avoid repeating undercounts seen in other emerging chronic conditions post-pandemic.