Readability is an almost century-old field that aims to match readers with texts based on reproducible tools. It has a long history, but it is remarkable that the methodology used to coin readability formulas hardly evolved until recently, when traditional readability studies came into contact with the field of Natural Language Processing (NLP). In this paper, we first briefly retrace the milestones in readability research before the advent of this new paradigm, that we propose to call ‘Artificial Intelligence (AI) readability’ since it results from the conjunction of two AI-connected domains: NLP and machine learning. We then describe in more detail some of the important studies within this new paradigm, before discussing three main issues of the field that need to be resolved in order to make significant advances and present interesting perspectives for future research.