Quality, transparency, and readability of artificial intelligence chatbots as sources of information on neonatal hyperbilirubinemia: a web-based cross-sectional comparative study

Background Neonatal hyperbilirubinemia is one of the most common clinical problems during the neonatal period. Parents often need to understand information on increased bilirubin levels, phototherapy, exchange transfusion, feeding, and follow-up. As artificial intelligence (AI) chatbots increasingly serve as supplementary channels for public medical information seeking, the quality, source transparency, and linguistic comprehensibility of chatbot-generated content may directly influence patient education. Objective To evaluate the quality, transparency, overall educational value, and readability of neonatal hyperbilirubinemia-related information generated by four mainstream AI chatbots, and to determine whether this information meets the recommended sixth-grade readability level for patient education materials. Methods This web-based cross-sectional comparative study identified core questions related to neonatal hyperbilirubinemia using MeSH and Google Trends. The questions were entered into ChatGPT 5.5, Gemini 3.5 Flash, Perplexity, and Copilot under standardized conditions. Information quality, completeness, transparency, and overall educational value were assessed using DISCERN, EQIP, JAMA benchmark criteria, and the Global Quality Scale (GQS). Readability was assessed using the Automated Readability Index (ARI), Gunning Fog Index (GFI), Flesch-Kincaid Grade Level (FKGL), Coleman-Liau Index (CL), Simple Measure of Gobbledygook (SMOG), and Flesch Reading Ease Score (FRES). Between-chatbot differences were analyzed using the Friedman test, followed by paired Wilcoxon signed-rank tests with Holm correction when overall differences were significant. Results Of 50 candidate search items initially identified, 22 were excluded and 28 core questions were included. Significant differences were observed among the four chatbots in DISCERN, EQIP, JAMA, and GQS scores (all P < 0.001). Copilot and Perplexity performed better overall on DISCERN, EQIP, and JAMA, but their absolute quality did not reach an excellent level. ChatGPT and Gemini showed particularly limited source transparency. All grade-level readability indices exceeded the sixth-grade threshold, and all FRES values were below 80, indicating that the generated texts were too difficult for general patient education. Conclusions AI chatbots can provide structured supplementary information on neonatal hyperbilirubinemia, but substantial between-model differences, insufficient source transparency, and excessive reading complexity remain common limitations. Future development should strengthen authoritative source tracing, guideline-version disclosure, content completeness review, and plain-language generation strategies for populations with lower health literacy. Clinical trial number Not applicable.