Generative AI clinician tools need to be built for pediatrics

Clinical takeaway: Ask whether a generative AI tool was validated in patients of the ages you treat, and remember that final responsibility for decisions stays with the clinician. 

AI that turns an exam room conversation into a visit note, or drafts replies to patient messages, is running in many hospital systems and practices. Pediatricians who use it care for patients whose diagnoses, treatments, and disease differentials change with age. But AI designed for general health care settings may not fit those workflows. 

Several medical societies have already issued guidance on generative AI. The American Academy of Pediatrics argues that pediatric care needs a framework of its own. Validation in everyday clinical use is still limited, it notes, and a data breach can follow a patient well into adulthood. The policy statement sets out that framework, with recommendations for developers, health care institutions, and regulators. 

The statement cites a 2024 evaluation on board examinations in which GPT-4 performed worse in pediatrics than in general surgery, internal medicine, and psychiatry. Developers should train tools on data that represents the full range of pediatric patients, the statement says, because adult data may fail to capture the physiological and developmental differences between children and adults. It also asks developers to validate performance across ages, developmental stages, races and ethnicities, genders, languages, and abilities, and to report those results openly. When a tool falls back on adult data because no pediatric data exist for a prompt, the output should say so. 

Health care institutions are asked to confirm that developers followed those validation practices before they select a tool, and to review how it monitors performance and reports errors. The statement also warns that publicly available large language models often default to settings that don't comply with the Health Insurance Portability and Accountability Act, so clinicians need to take care with what they enter. Current clinical use requires a human in the loop, the statement says, and any move toward less supervision should wait for rigorous validation and regulatory approval. 

Regulators are asked to obtain evidence of safety and efficacy in children before approving generative AI systems for pediatric use. The statement also calls for postmarket surveillance that accounts for how model performance can drift as the underlying systems are updated. 

The statement came from the academy's Council on Clinical Information Technology and Section on Innovation in Therapeutics and Technology, whose authors reviewed and interpreted the relevant literature. It covers pediatricians' use of generative AI in clinical care, not use by children, adolescents, or families. 

FDA has proposed regulatory frameworks for generative AI, but the statement notes they are still in development. It adds that the vast majority of FDA-cleared algorithms have been static ones that don't vary over time the way generative systems do. The academy is also preparing separate guidance on direct use of AI by families, which this statement does not cover. 

"Generative AI is already providing real value to pediatricians, and it's improving at an incredible pace," said R. Brandon Hunter, MD, an author of the statement and a pediatric intensivist at Texas Children's Hospital. "But what makes this moment in healthcare so interesting and unusual is that adoption is often moving faster than the evidence on how to use these tools effectively is being produced. We hope this statement gives pediatricians a framework for thinking about AI implementation as that evidence catches up." 

Source: Suresh S, et al. (2026 Oct 3) Pediatrics. Recommendations for the Development and Implementation of Generative Artificial Intelligence Tools in Pediatric Clinical Care: Policy Statement