Answers are grounded in this paper and cite their sources.
Purpose To compare the reading level of paediatric health guidance, parents' own questions and artificial intelligence (AI) chatbot responses to those questions, and to test whether chatbots adapt to the reading level of the person asking. Methods Cross-sectional study, reported following the Chatbot Assessment Reporting Tool (CHART), of 18,392 guideline pages (17,516 family-facing, 876 clinician-facing) from 86 organisations, 1,000 parent-authored questions sampled from 23,477 public forum posts, and 5,000 responses to those questions from five commercially deployed chatbots. Flesch–Kincaid Grade Level (FKGL) was the primary outcome, with six secondary metrics. In a sub-study, 100 questions were resubmitted with an instruction to match the reader's level or to answer at sixth-grade level (1,500 responses). Results Median FKGL was 12.51 for clinician-facing guidance, 9.08 for family-facing guidance, 8.37 for chatbot responses and 6.15 for parents' questions. The American Medical Association sixth-grade standard was exceeded by 99.3% of clinician-facing pages, 92.6% of family-facing pages and 89.2% of chatbot responses. Every chatbot answered above the reading level of the question asked (median difference + 1.21 to + 3.57 grades). Given an explicit sixth-grade instruction, the median gap narrowed in every model and three of five answered below the reader's level. Conclusion Paediatric health information from institutions and chatbots alike is written above the level of its readers. Chatbots do not adapt to the reader by default, but can when instructed.
No takes yet. Share an insight, caveat, or question.
Cunningham et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: