<p>As programs and applications for artificial intelligence (AI) become more ubiquitous in educational settings, conversations about accessibility of AI are needed. Research and practice can still use unclear language about access, accessibility, and related terms such as assistive technology. This review focused on identifying a small group of research studies that were attempting to consider the accessibility of the AI tools themselves—rather than use AI as the accessibility feature. Those papers were diffractively read for insights about what considerations are needed to ensure that accessibility conversations for AI include a broad range of understandings about technical accessibility, critical theories, and ethics. The implications of this review include the need for institutions to consider the differences between using AI to do <i>with students,</i> rather than <i>do for</i> or <i>do to</i> them in educational settings<i>.</i> Positioning AI for <i>doing with</i> enables stronger conversations about what it means to make AI accessible.</p>