WED-011 - Advancing Health Literacy Through AI: Developing and Testing a Custom GPT Model

Brief Abstract Summary: Explore how a custom GPT model can improve health communication by assisting health communicators in creating plain language materials that are easier for all patients to understand and use. This poster explores how AI editing tools can help simplify medical information, eliminate jargon, and ensure readability at an appropriate level, making materials more accessible for all patients. Participants will see examples of how the tool helps produce health materials that prioritize readability and clarity, meeting standards set by the Patient Education Materials Assessment Tool (PEMAT) to make information more understandable and usable for diverse audiences. Attendees will gain insight into creating AI tools that meet health literacy needs and promote clearer, more actionable communication across diverse patient populations. This poster is for anyone interested in using AI in health communication.

Detailed abstract description:

Introduction: Effective health communication plays a vital role in promoting health literacy, ensuring that patients can make informed decisions about their health . Writing and editing in plain language is a specialized skill that requires practices and training, and not everyone has the training or experience to do it effectively. However, we propose that AI can be trained to integrate plain language best practices, making it easier for writers to create clear and accessible information.

Methods: Our poster outlines the process we used to build a plain language knowledge repository for our custom AI tool. We began by researching and practicing prompt engineering by attending courses and seminars about how to effectively craft AI inputs. Next, we created a custom GPT model by adding specific instructions, uploading our plain language checklists, and feeding it examples of texts that met our standards. Finally, we tested the AI’s output by feeding it original patient education materials written by clinicians and asked it to apply plain language best practices, using prompt engineering techniques to help it best access and use the repository.

Results: Participants will see how the tool can be applied to real-world texts, with specific examples of patient education materials that illustrate its effectiveness. We will share results from editing examples produced by the AI model using standard processes to assess readability, understandability, and actionability, as evaluated through the Patient Education Material Assessment Tool (PEMAT).
Conclusion and Limitations: This poster will focus on responsible AI use and highlight how we created a customized tool to meet health literacy standards and create clear, actionable plain language writing. Moreover, we will present limitations of AI tools, including the need for manual content checks, and share strategies to address these challenges.