AI and book indexing: trajectory data

While artificial intelligence (AI) can be a highly useful tool in some situations and with some highly defined problem spaces, book indexing does not appear to be one of them. This paper supplements our previously published white papers by comparing the effectiveness of large language model AI chatbots, across multiple generations, at indexing a book or book chapter. Our earlier white papers indicated that AI-generated indexes fail to reflect subtopics and related topics, preventing readers from having appropriate access to all indexable material; that AIs under-index, failing to pick up on significantly discussed terms; that at the same time they also over-index, cluttering the index with irrelevant entries and redundant subheadings; and that AI-generated indexes also fail a standard copy-editor’s test for accuracy. The current paper indicates that AIs have not improved significantly on any measure.