Abstract This study uses machine learning techniques to model patent claim clarity and analyze how clarity relates to important patent policy objectives. Specifically, machine learning models are trained on a dataset of over 600,000 U.S. patent applications that were (or were not) rejected for indefiniteness, a proxy for claim clarity, using features based on the linguistic attributes of each application. The model is then applied to over 2 million issued patents and their corresponding applications, deriving estimates of the clarity of each patent's claim set at application and issuance. First, the properties of claim clarity and its relationship with the patent examination process are studied. Wordiness and repetitiveness corresponds to reduced clarity, whereas more descriptiveness whereas clearer claims tend to be more descriptive. Clarity also changes during patent examination, indicating that patent office policies may affect claim clarity. Next, the relationship between claim clarity and cumulative innovation is studied. Clear patents are found to receive more citations by applicants of unrelated future patents, a key indicator of cumulative innovation. However, unclear patents tend to receive more examiner citations, particularly in later years, and the technological relevance of examiner citations also tends to decline over time. This raises important questions about the role of late-stage examiner citations in the patent examination process, which are framed for future research. Finally, this study evaluates the impact of the U.S. Supreme Court's Nautilus v. Biosig decision, which sought to improve patent claim clarity. A difference-in-difference analysis of applications examined under the old versus new standard is conducted to evaluate the causal effect of Nautilus on the claims of patents filed under the old standard but examined under the new standard. This reveals a significant improvement in patent claim clarity post-Nautilus.