Abstract
This chapter explores the application of ChatGPT in reading test development and validation through two case studies. Study 1 evaluates ChatGPT’s ability to estimate item difficulty and discrimination for a reading comprehension test without response data from test takers. Results show that ChatGPT estimated item difficulty with moderate agreement with empirical data, though it struggled with estimating item discrimination. Study 2 investigates ChatGPT's capacity to answer multiple-choice reading items without access to passages, revealing correct response rates that notably surpassed two human participants. These findings indicate ChatGPT’s potential as a tool for facilitating reading test development and validation, particularly in providing early-stage validity evidence when pilot data are unavailable. Future studies should explore diverse test formats and refine AI methodologies to enhance their utility in measurement practice.