An analysis of AI platforms: Can AI provide genetics education?

Abstract

With the introduction, free access, and proliferation of various artificial intelligence (AI) in unique platforms, patients have access to more summarized information than ever. Our study aimed to explore the use of AI chatbots to improve patients' understanding of genetic testing. Thirty‐nine frequently asked questions (FAQs) on genetic testing, including generic, BRCA‐related, and Lynch syndrome‐based questions, were asked to five AI chatbots (ChatGPT 3.5, Google Gemini, Reddit Answers, Bootcamp, and DeepSeek). Answers were individually assessed in a blind, randomized survey by two gynecologic oncologists and two genetic counselors on a 4‐point scale (1: accurate and comprehensive, 2: accurate but inadequate, 3: incomplete, partially inaccurate, and/or outdated, 4: completely inaccurate) and statistically evaluated with ANOVA and Tukey testing. A Flesch–Kincaid Grade Level (FKGL) readability score and word count for each answer were obtained. Overall ratings show that no single AI platform outperforms the others in accuracy or readability, though Reddit Answers was notably worse. On average, ChatGPT 3.5, Google Gemini, Reddit Answers, Bootcamp, and DeepSeek received scores of 1.5, 1.6, 2.4, 1.6, and 1.4, respectively. The FKGL readability scores of answers across all AI platforms are consistently higher than the reading level of the average American (ChatGPT 3.5: 13.5, Google Gemini: 13.3, Reddit Answers: 12.1, Bootcamp: 14.6, DeepSeek: 14.2). Although AI chatbots have the potential to be a great resource for patients seeking genetic testing, especially for foundational genetic knowledge, this study highlights the importance of genetic counseling and education by healthcare professionals. Because genetic testing is unique to each patient, the answers produced by AI chatbots are not always applicable to each individual.