Evaluating the constructiveness and professional tone of AI-generated corrective feedback for medical learners

In a blinded comparative evaluation reported in BMC Medical Education, ChatGPT-generated corrective feedback scripts received higher combined median ratings than educator-authored scripts across four assessed domains and were preferred by raters for 78% of 18 learner-deficiency scenarios. The study focused on communication characteristics rather than the factual accuracy, educational effects or clinical applicability of the feedback.

Six pediatric medical education leaders independently rated narrative feedback scripts for 18 standardized learner-deficiency scenarios. For each scenario, one script had been authored by an educator and the other had been generated by ChatGPT without editing. Raters used a three-point scale, from 0 for “No” to 2 for “Yes,” to assess whether scripts were behavior-centric, balanced between positive and negative feedback, non-judgmental and focused. The researchers compared median domain scores and an overall median score for the two groups of scripts.

The abstract reports that, when results from all four domains were combined, the ChatGPT-generated scripts had higher median scores and increased odds of receiving higher domain scores than the educator-authored scripts. Raters also preferred the ChatGPT-generated scripts for 78% of the scenarios. These findings represent the ratings and preferences of six pediatric medical education leaders evaluating standardized cases; they do not establish that AI-generated feedback improves learner performance or produces better educational outcomes.

The authors described the findings as suggesting that AI-generated scripts may have potential as an adjunct to faculty feedback and that AI could function as a supportive drafting tool. They also stated that further study is required before conclusions about real-world utility can be made. Because the available report evaluates constructiveness and professional tone across four communication domains, it does not address feedback accuracy, educational impact or clinical applicability, and the abstract does not provide details supporting wider generalizability.

Subject of Research: Blinded comparison of ChatGPT-generated and educator-authored corrective feedback scripts for pediatric medical learners

Article Title: Evaluating the constructiveness and professional tone of AI-generated corrective feedback for medical learners

Article References: Desselle, B., Simon, E., Prudhomme, A., Mumphrey, C., Reilly, L., Huntwork, M., Sarkar, S., Hescock, G., Fang, Z., Patrick Esteve, J., Gajewski, K., & Creel, A. (2026). Evaluating the constructiveness and professional tone of AI-generated corrective feedback for medical learners. BMC Medical Education. https://doi.org/10.1186/s12909-026-10547-6

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10547-6

Keywords: medical education, ChatGPT, generative AI, corrective feedback, medical learners, pediatric education, educator feedback, constructiveness, professional tone, blinded evaluation, Evaluating, professional