Artificial Intelligence (AI) is becoming widespread. There are many benefits associated with AI, but it’s adoption brings challenges relating to fairness, bias, and transparency. Such issues are particularly hard to address because conventions that highlight when an AI is present, how it works, and the consequences of using are not yet established: AI has a legibility problem. Design-led research can play a key role in exploring this challenge. Applying Research through Design (RtD) this paper explores AI legibility in three ways: (1) explaining why it makes sense to address AI legibility with design; (2) the presentation of prototypical icons designed to enhance AI legibility; (3) experimenting with how the icons may be used in the context of signage relating to potential applications of AI. Via these three lenses the paper argues that design’s role in improving AI legibility is critical.