Large language models as a source of health information: Are they patient-centered? A longitudinal analysis

Individuals often turn to search engines like Google Search as their first source of health information. However, large language models (LLMs) are positioned to become another destination for those seeking medical information. Consequently, the readability of these materials becomes an important factor in ensuring their effectiveness in promoting health literacy, given that the average American reads at the eighth-grade level.1,2 We looked at the readability and word count of information produced by four publicly available LLMs – ChatGPT-3.5, ChatGPT-4, Google Bard (Gemini), and Microsoft Bing (set to precise) – across 4 months.