Recent developments in AI chat are sending shockwaves through the language teaching community, both with short-term challenges of instructing students when and how to use this technology and as a longer-term existential threat to the teaching vocation. On the other hand, this same technology presents an opportunity for the automatic production of compelling input, not only in English but potentially for many other languages. Critical to providing suitable input is determining the level of readability, for example measured in YL (Yomiyasusa Level), which is based on impressions of difficulty by readers in Japan. This presentation reports on research into machine learning techniques used to estimate YL using the Coh-metrix analysis tool, Lasso linear regression and grid search cross-validation. The model predicted YL with a strong correlation of .91, significantly better than the Flesch Reading index. The results suggest that the developed model is a promising tool for predicting YL.