Proficiency of artificial intelligence in addressing patient frequently asked questions on oral lichen planus.

OBJECTIVES

To investigate the proficiency of artificial intelligence (ChatGPT-4 and DeepSeek-v3) in responding to frequently asked questions (FAQs) related to oral lichen planus (OLP).

STUDY

DESIGN

Twenty-three OLP-related FAQs were adapted from known sources. Responses generated by ChatGPT-4 and DeepSeek-v3 were assessed by expert panels across 5 domains: accuracy, comprehensibility, logic, fairness, and simplicity.

RESULTS

DeepSeek-v3 outperformed ChatGPT-4 in comprehensiveness (4.65 ± 0.37 vs 4.04 ± 0.54; P < .001), logic (4.41 ± 0.32 vs 4.03 ± 0.35; P < .001), and fairness (4.34 ± 0.33 vs 4.03 ± 0.27; P < .001). ChatGPT-4 scored higher in simplicity (4.25 ± 0.31 vs 4.04 ± 0.28; P = .034), while DeepSeek-v3 had a modest advantage in accuracy (4.49 ± 0.38 vs 4.22 ± 0.34; P = .025). DeepSeek-v3 performed better in comprehensiveness and logic for "Causes" and "Symptoms," and in accuracy for "Risk & Complications." ChatGPT-4's simplicity was most evident in "Causes" and "Symptoms." In "Treatment & Management," DeepSeek-v3 excelled in logic and comprehensiveness, while ChatGPT-4 showed greater simplicity and occasional accuracy.

CONCLUSION

ChatGPT-4 provides concise, accessible information, whereas DeepSeek-v3 offers broader, more detailed responses. Combining both models may improve patient understanding of OLP.