The accessibility tree, a structure built for screen readers, now dictates whether AI agents like ChatGPT and Gemini can access web pages. This semantic structure, which browsers have used for decades, is crucial for AI understanding. Automated traffic has surpassed human traffic, making AI a significant portion of web visitors. These agents don't see visual design; they rely on well-labeled structure. Many websites lack this crucial structural information, rendering them inaccessible to these new visitors. Vendors are splitting on how AI agents ingest page data, with some prioritizing screenshots and others the accessibility tree. However, even vision-first approaches are increasingly leveraging structured data. Transactional and dynamic interfaces are most affected by poor structure, mirroring issues for keyboard-only users. Studies show AI agents perform significantly worse when constrained to accessibility-tree-dependent navigation. Building for agent legibility is distinct from endorsing their actions or the potential for content to be "lifted" without user clicks. The foundational steps to improve this include using native HTML elements, logical headings, and landmarks. Correctly naming, defining roles, and updating states for interactive elements are essential. ARIA should be used judiciously, prioritizing native HTML semantics. Image alt text and proper form field labeling are also critical for AI comprehension. Finally, providing feedback through state announcements and maintaining stable layouts ensures agents can navigate and interact effectively.