Generative AI (Gen AI) has the potential to offer a more intuitive, meaningful discovery experience. However, it also has the potential to obfuscate and distort information. Given the growing number of AI tools emerging, libraries must balance the need to make discovery both as easy and as accurate as possible. In this presentation, we describe how we developed an approach to assessing user-focused AI tools and our assessment of the Primo Research Assistant. We will discuss common issues with AI, testing and assessment strategies, and what we found in our evaluation of the Research Assistant.