A study published in *Retina* found that ChatGPT-5 overpredicts success after macular hole surgery. Investigators led by Aya Wattad, MD, at Tel Aviv University evaluated 50 eyes of 50 patients who underwent pars plana vitrectomy for full-thickness macular hole between 2021 and 2024. ChatGPT-5 predicted anatomical closure in 49 of 50 eyes, achieving 90% accuracy, ahead of two senior retina specialists' 72% to 86%, but was "blind to the eyes that did not" close. At 12 months, actual anatomical closure occurred in 44 of 50 eyes (88%). For functional outcomes, overall best-corrected visual acuity (BCVA) prediction accuracy was 66% for ChatGPT-5 versus 42% to 44% for specialists, but the model failed entirely when vision worsened (0% accuracy). The mean BCVA prediction error was 11.4 ± 10.8 letters. The authors concluded that the model's outputs "require cautious interpretation to avoid misleading confidence" and that AI-generated prognoses should serve only as supportive information.