Related Experiment Video
Updated: Aug 27, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Prompting generative artificial intelligence to create plain language patient education materials
Sean M Frey1, Andrea Milne Wenderlich1, Crystal Craig1
1Department of Pediatrics, University of Rochester School of Medicine & Dentistry, Rochester, NY, USA.
Objectives:
Generative artificial intelligence (AI) holds promise for creating patient education materials (PEM). However, AI-produced content is often difficult to read. Prompts for AI to write at a lower reading level either fail to meet recommended targets (≤6th grade) or demonstrate inconsistent results. We aimed to determine whether a new AI prompt for plain language would create PEMs that are easier to read, understand, and act upon.
Methods:
We analyzed PEMs from two websites (HealthyChildren.org, UpToDate) and two generative AI programs (ChatGPT, Gemini). For each AI program, we used a basic prompt and an advanced prompt to generate separate documents. From these sources, we collected six sets of PEMs about five acute pediatric conditions, evaluated them using a panel of readability indices, and used ANOVA to compare average readability scores between sources. Two pediatricians assessed understandability and actionability using the Patient Education Materials Assessment Tool (PEMAT). We calculated mean PEMAT scores, with values > 70% considered understandable/actionable.
Results:
ChatGPT (advanced prompt) created PEM with the lowest mean grade reading level (5.8; SD 0.84) and highest PEMAT scores (91% understandable, 84% actionable). Documents from generative AI using the advanced prompt and UpToDate had mean readability scores ≤ 6th grade, with significantly better readability than AI-generated documents using the basic prompt or HealthyChildren.org. All sources except HealthyChildren.org had mean understandability > 70%; AIs using the advanced prompt and UpToDate exceeded the actionability threshold.
Conclusions:
A plain-language prompt for generative AI created PEMs that were readable, understandable, and actionable, with comparable results across two AI programs.
Practice Implications:
Healthcare providers can use this new plain-language prompt for generative AI programs to quickly create free, patient-friendly PEM about acute healthcare topics. As these AI programs were not trained to provide medical guidance, medical professionals should continue to review PEM prior to sharing with patients.
Related Concept Videos
Patient-centered Care
Non-equilibrium in the Cell
Health Literacy
Documentation in Long-Term and Home Healthcare Setting
Long-Term Care Facilities
Nursing Process for Patient and Caregiver Teaching II: Planning and Implementation
Nursing Process for Patient and Caregiver Teaching I: Assessment and Diagnosis
It is critical to determine the patient's learning needs during the assessment. Determination of learning needs compounds data from the...