Background/Objectives: Written discharge explanations are crucial for patient understanding and safety in emergency eye care, yet their tone and clarity vary. Large language models (LLMs, artificial intelligence systems trained to generate human-like text) can produce patient-friendly materials, but direct, blinded comparisons with clinician-written texts remain scarce. This study compared patient perceptions of a routine clinician-written discharge text and a GPT-5-generated explanation, where GPT-5 (OpenAI) is a state-of-the-art LLM, based on the same clinical facts in emergency ophthalmology. The primary objective was empathy; secondary outcomes included clarity, detail, usefulness, trust, satisfaction, and intention to follow advice. Methods: We conducted a prospective, single-blind, within-subject study in the Emergency Ophthalmology Unit of the University Hospital Centre Split, Croatia. Adults (n = 129) read two standardized texts (clinician-written vs. GPT-5-generated), presented in identical format and in randomized order under masking. Each participant rated both on Likert scales with 1–5 points. Paired comparisons used Wilcoxon signed-rank tests with effect sizes, and secondary outcomes were adjusted using the Benjamini–Hochberg false discovery rate. Results: Empathy ratings were lower for the GPT-5-generated text than for the clinician-written text (means 3.97 vs. 4.30; mean difference −0.33; 95% CI −0.44 to −0.22; p < 0.001). After correcting for multiple comparisons, no secondary outcome differed significantly between sources. Preferences were evenly split (47.8% preferred GPT-5 among those expressing a preference). Conclusions: In emergency ophthalmology, GPT-5-generated explanations approached clinician-written materials on most perceived attributes but were rated less empathic. A structured, human-in-the-loop workflow—in which LLM-generated drafts are reviewed and tailored by clinicians—appears prudent for safe, patient-centered deployment.

Accurate documentation is critical in ophthalmology, yet clinical notes often contain subtle errors that can affect decision-making. This study prospectively compared contemporary large language models (LLMs) for detecting clinically salient errors in emergency ophthalmology encounter notes and generating actionable corrections. 129 de-identified notes, each seeded with a predefined target error, were independently audited by four LLMs (o3 (OpenAI, closed-source), DeepSeek-v3-r1 (Deepseek, open-source), MedGemma-27B (Google, open-source), and GPT-4o (OpenAI, closed-source)) using a standardized prompt. Two masked ophthalmologists graded error localization, relevance of additional issues, and overall recommendation quality, with within-case analyses applying appropriate nonparametric tests. Performance varied significantly across models (Cochran’s Q = 71.13, p = 2.44 × 10⁻¹⁵). o3 achieved the highest error localization accuracy at 95.7% (95% CI, 89.5–98.8), followed by DeepSeek-v3-r1 (90.3%), MedGemma-27b (80.9%), and GPT-4o (53.2%). Ordinal outcomes similarly favored o3 and DeepSeek-v3-r1 (both p < 10⁻⁹ vs. GPT-4o), with mean recommendation quality scores of 3.35, 3.05, 2.54, and 2.11, respectively. These findings demonstrate that LLMs can serve as accurate “second-eyes” for ophthalmology documentation. A proprietary model led on all metrics, while a strong open-source alternative approached its performance, offering potential for privacy-preserving on-premise deployment. Clinical translation will require oversight, workflow integration, and careful attention to ethical considerations.