Google AI director outlines new content playbook

Addy Osmani, a director of engineering at Google Cloud AI, printed new steerage on Agentic Engine Optimization (AEO), a mannequin for making content material usable by AI brokers.

He positioned this AEO (to not be confused with Reply Engine Optimization) as parallel to SEO, constructed for methods that fetch, parse, and act on content material autonomously.

What he’s seeing. AI brokers collapse multi-step searching right into a single request. They don’t scroll, click on, or interact with UI — they extract what they want immediately. That makes most conventional engagement metrics irrelevant.

The token downside. Osmani highlighted token limits as a core constraint shaping content material efficiency. Massive pages can exceed an agent’s context window, inflicting:

• Truncated data.
• Skipped pages.
• Hallucinated outputs.

His takeaway: token depend is now a main optimization metric.

Content material wants to vary. Osmani really helpful restructuring content material for a way brokers learn:

• Put solutions early (ideally throughout the first ~500 tokens).
• Preserve pages compact and centered.
• Keep away from lengthy preambles and buried insights. (Brokers have “restricted persistence” for this, he famous.)

Markdown over HTML. He additionally really helpful serving clear Markdown alongside conventional pages.

• Markdown reduces noise from navigation, scripts, and structure, making content material simpler and cheaper for brokers to parse.
• This contains making .md variations straight accessible and discoverable.

Discovery and construction. Osmani pointed to rising patterns for serving to brokers discover and use content material:

• llms.txt as a structured index of documentation.
• ability.md recordsdata to outline capabilities.
• AGENTS.md as a machine-readable entry level for codebases.

These act as shortcuts for brokers deciding what to learn and use.

Why we care. This provides a brand new optimization layer alongside search engine marketing. If brokers can’t effectively parse your content material — resulting from token limits, construction, or format — they could skip, truncate, or misread it. That straight impacts whether or not your content material is used, cited, or acted on in AI-powered experiences.

Between the strains. To be clear, the kind of AEO Osmani mentioned in his article is unrelated to Google Search or natural search rating. Of notice, Google’s John Mueller recommended against markdown pages and Google doesn’t use the llms.txt file.

• Osmani’s article highlights how AI methods work together with the online and what “optimized” content material could seem like in that atmosphere.
• AEO shifts the objective from driving visits to enabling profitable outcomes inside AI workflows.

The article. Agentic Engine Optimization (AEO)

Dig deeper: