Introduction: The ablation index (AI) standardizes lesions in radiofreqency ablation (RF) of atrial fibrillation (AF). High-power, short-duration ablation (HPSD) shortens procedures, but AI is histologically unvalidated in HPSD. We evaluated esophageal safety with voltage-adapted AF ablation using an optimized, AI-guided HPSD (AI-HPSD) protocol. Methods: Consecutive AF patients undergoing AI-HPSD at 50W (AI=400-450 for posterior left atrium (LA); AI ≥500 anteriorly) were compared with a recent AF cohort who underwent AI-guided low-power, long-duration ablation (AI-LPLD). All participated in our ablation registry. Esophageal endoscopy was performed 1-3 days post-ablation. Posterior wall lesion characteristics and endoscopically-detected esophageal lesions (EDEL) were compared. Results: AI-LPLD (n=100) and AI-HPSD (n=100) groups had similar baselines, except AI-LPLD had slightly larger LA-area. VISITAG numbers were comparable, but total posterior wall RF time was shorter with AI-HPSD versus AI-LPLD (4.5±1.6 versus 11.9±5.1 s, p<0.001). Mean AI and mean impedance drop were higher in AI-HPSD (449±17 versus 401±39; p<0.001 and 8.8±2.3 versus 7.8±2.5; p<0.001). AI standard deviation was lower with AI-HPSD (30±15 versus 49±19; p<0.001). Although Max AI values were higher in AI-HPSD (517±46 versus 496±55; p=0.003), maximum impedance drop did not differ significantly. EDEL rate was comparably low at 2% (AI-HPSD) and 4% (AI-LPLD), but Max AI >520 occurred more with AI-HPSD (3.5±5.5 versus 2.5±5.2; p=0.048). Conclusion: In routine practice, optimized AI-HPSD at 50W with posterior LA target AI =400-450 was not associated with more EDEL. However, a signal for more unacceptable Max AI values existed. We recommend limiting AI to 400 and developing an automatic, AI-guided ablation stop.