Artificial Intelligence in Endodontics: A Narrative Review and Critical Synthesis of Diagnostic, Predictive, and Operational Applications

Background/Objectives: Artificial intelligence (AI) research in endodontics has grown quickly, but the strength of the evidence differs between applications. This narrative review with structured evidence mapping examined diagnostic imaging, canal morphology, outcome prediction, large language models (LLMs), robotics and augmented reality, and operational uses such as case-difficulty assessment, triage and scheduling. Methods: Two anchor reviews were combined with a documented PubMed search to 30 September 2026, and studies were classified by design, sample, reference standard, validation, and evidence of clinical utility. Results: Periapical lesion detection has the largest evidence base: one meta-analysis reported pooled sensitivity of 0.925 and specificity of 0.852, although most included studies were at risk of bias and none was prospective. Fracture detection relies mainly on ex vivo data and is sensitive to image resolution. Outcome models showed modest to moderate discrimination in single-centre data. LLM studies reported 40% to 100% correct, appropriate, or concordant responses across different benchmarks and scoring rules, and outputs varied on repeated querying. Difficulty-assessment and scheduling tools have appeared since 2020 but have not been evaluated prospectively in clinical workflow. Conclusions: Technical performance has not yet been shown to translate into clinical benefit, and AI in endodontics should currently be used only as clinician-supervised decision support.