The Readability of Generative AI Policies: An Empirical Analysis of OpenAI’s Informational Documents

Abstract

In recent years, media attention has focused on artificial intelligence, particularly on chatbot services and generative intelligence. ChatGPT, created by OpenAI, was one of the earliest online tools and rapidly gained popularity. Users are indeed exposed to a service with privacy notifications and conditions of use that control service use and personal data collecting, limiting misuses. However, it is crucial that these materials must be usable to ensure ethical use of this technology. This exploratory single-provider case study assesses textual properties of 11 English-language OpenAI informational documents. Specifically, several readability indices have been computed (i.e., Gunning Fog Index, Coleman–Liau Index, Flesch–Kincaid Grade Level, Automated Readability Index, Simple Measure of Gobbledygook, and Flesch Reading Ease), along with lexical density and estimated silent and spoken reading time. A comparison between the properties of the documents intended for general consumers (n = 4) and professional/business-to-business (n = 7) was performed. Results showed that most documents exhibit high structural linguistic complexity. Group-level tests did not identify significant differences between the two audience groups after correction for multiple comparisons, whereas the readability indices were strongly intercorrelated. Document length was not systematically associated with readability scores, and its relationship with estimated reading time was deterministic because reading time was calculated from word count. This study establishes a descriptive baseline for provider’s policy complexity. The findings identify potential structural and temporal barriers but do not demonstrate actual comprehension, accessibility, legal compliance, or informed consent.