Addy Osmani, a director of engineering at Google Cloud AI, has shared new recommendations on Agentic Engine Optimization (AEO), a framework for making content more accessible and useful to AI agents. He describes this version of AEO (distinct from Answer Engine Optimization) as a counterpart to SEO, but tailored for autonomous systems that retrieve, interpret, and act on content without human interaction. His observation: AI agents compress what used to be multi-step browsing into a single query. They don’t scroll, click, or interact with interfaces — they immediately extract the information they need. As a result, many classic engagement metrics lose their relevance.
The token constraint. Osmani emphasized that token limits are now a fundamental factor influencing how content performs. When pages are too large, they can exceed an agent’s context window, which can lead to: Information being cut off. Entire pages being skipped. Hallucinated or incorrect responses. His conclusion: token count has effectively become a key optimization metric.
Content must adapt. Osmani advised rethinking content structure around how agents actually consume it: Surface answers early (ideally within the first ~500 tokens). Keep pages concise and tightly scoped. Minimize long introductions and avoid hiding key insights deep in the text. (Agents have “limited patience” for this, he noted.)
Markdown vs. HTML. He further suggested offering clean Markdown versions alongside standard web pages. Markdown strips away noise from navigation, scripts, and layout, making it simpler and cheaper for agents to parse. This includes exposing .md files in a way that they’re directly accessible and easy to discover.
Discovery and information architecture. Osmani highlighted new conventions emerging to help agents locate and interpret content, such as: llms.txt as a structured index for documentation. skill.md files to describe capabilities. AGENTS.md as a machine-readable entry point for repositories. These serve as shortcuts that guide agents on what to read and where to start…