Google AI Director Publishes Agentic Engine Optimization Guidance

This feels like one of those moments when a niche idea starts turning into a real shift in how content gets built and published online.

Addy Osmani, Director of Engineering at Google Cloud AI, has published new guidance around what he calls Agentic Engine Optimization, or AEO. His framework is not aimed at traditional search engines or even standard AI answer engines. It focuses on something more specific: AI agents that read documentation, repositories, and product content directly in order to take action.

That distinction matters.

For years, most teams have optimized for humans first and search engines second. Osmani’s guidance suggests a third audience is now here in a serious way: autonomous or semi-autonomous AI agents like Claude Code, Cursor, Aider, and similar tools that read docs, call APIs, inspect repos, and execute workflows without behaving anything like a human visitor.

Why this guidance is getting attention

The April 2026 post gained traction quickly because it does not read like vague AI futurism. It is practical. It reflects something many developer-focused brands are already seeing: AI agent traffic is rising, but it is often undercounted, under-measured, and poorly supported.

Osmani’s core idea is straightforward: if AI agents are increasingly acting as the users of technical content, then that content needs to be structured so agents can consume it reliably.

He puts it plainly: token count is now a first-class documentation metric.

That line helps explain why the guidance is resonating. The question is no longer just whether content is helpful or ranks well. It is whether an AI agent can parse it efficiently, stay within context limits, and finish a task without hallucinating, skipping steps, or failing silently.

What AEO actually means

The easiest way to understand AEO is to separate it from the other acronyms it keeps getting grouped with.

• SEO is about visibility in traditional search results.
• GEO, or answer-engine optimization, is generally about being cited or surfaced in AI summaries and answer engines.
• AEO, as Osmani frames it, is about making content usable for agents that need to perform actions, not just generate summaries.

So the target is not only a chatbot citation. It is an agent trying to install a package, authenticate to an API, navigate a repository, or complete a coding task using your documentation.

This matters most for developer ecosystems, SaaS platforms, API companies, and technical publishers. If content is too long, too vague, poorly structured, or buried behind interfaces designed only for human browsing behavior, an agent may fail before it ever reaches the useful part.

The key recommendations in Osmani’s playbook

What stands out most is how operational the guidance is. This is not theory. It is a checklist for making content more machine-efficient.

1. Reduce token bloat

Quick-start guides should ideally stay under 15,000 tokens. That recommendation reflects a basic reality: agents do not browse patiently. They ingest content in bounded contexts.

If a guide stretches across tens of thousands of tokens, the odds go up that the agent truncates it, misses important steps, or produces flawed output. Long-form content can still work well for humans. For agents, excessive length becomes a usability problem.

2. Publish agent-friendly files

Osmani points to the rise of files such as:

• llms.txt as a lightweight index for AI systems
• skill.md to describe capabilities and action patterns
• AGENTS.md as a structured repository entry point

These are emerging conventions that help agents understand where to look, what matters, and how to act. They are not the same as SEO metadata, and they are not mainly about Google rankings. They are functional orientation layers for AI systems.

3. Serve content in Markdown where useful

Markdown is easier for many agents to parse than heavy, interactive web layouts. Google Search has made clear that this does not directly affect organic rankings, but that is beside the point here.

This is not about pleasing Google’s crawler. It is about helping agents consume content with less friction.

4. Audit for agent accessibility

Osmani also references tooling like agentic-seo, which works almost like a Lighthouse-style audit for AI agent usability. This category will likely grow quickly. Once brands see agents as a meaningful source of downstream action, they will want diagnostics that show whether their docs are actually machine-readable in practice.

5. Design for dual audiences

This may be the most important takeaway: the best content now needs to serve both humans and machines.

That does not mean writing robotic copy. It means structuring content so people can scan it and agents can execute from it. Good headings, concise summaries, explicit steps, capability definitions, and clear documentation architecture all become more valuable.

Why this changes content strategy

This guidance points to a shift from optimizing for visits to optimizing for outcomes inside AI workflows.

That is a very different mindset.

A human reader might tolerate some narrative, brand flourish, or exploratory structure. An AI agent usually will not. It wants the answer early, the steps clearly, and the environment to be predictable. If it cannot find those things quickly, it may improvise, and that is where hallucinations, broken integrations, and abandoned workflows begin.

For brands, the implication is significant. Documentation, help centers, product pages, and resource hubs are no longer just educational assets. They are becoming operational surfaces for AI systems.

That turns structure, brevity, and machine-oriented clarity from a nice extra into a competitive advantage.

The important caveat: this is not traditional Google Search advice

One reason the discussion has become muddled is the overlap in acronyms.

Google Search Advocate John Mueller has reportedly clarified that things like llms.txt and Markdown pages are not ranking factors for Google Search. So if anyone treats Osmani’s post as official SEO guidance for organic rankings, they are reading it the wrong way.

The better interpretation is this: Osmani is describing how to prepare content for an emerging agent ecosystem, not rewriting the Google Search playbook.

That nuance matters because it keeps teams focused on the right goal. If you are a publisher, SaaS company, or technical brand, the opportunity is not “rank better on Google because of AEO.” It is “be easier for AI agents to understand, trust, and use.”

What happens next

This conversation is likely to expand quickly.

The standards Osmani references—llms.txt, skill.md, and AGENTS.md—are still early, but the direction is clear. Documentation and product content are moving toward a model where they need to function as agent-readable infrastructure.

That will affect more than developer docs. Ecommerce, support content, onboarding flows, and product knowledge bases are all likely to move in this direction. As agents become more capable, content that is hard for them to parse will effectively become less visible in the places that matter.

And that may be the real headline: optimization is no longer only about being found. It is about being understood well enough for action.

FAQ

What is Agentic Engine Optimization?

Agentic Engine Optimization, or AEO, is the practice of structuring content so AI agents can read it, understand it, and use it to complete actions such as navigating documentation, calling APIs, or executing workflows.

How is AEO different from SEO?

SEO focuses on visibility in search results. AEO focuses on usability for AI agents that need to act on content, not just index it or summarize it.

Does AEO improve Google rankings?

Not directly. Guidance around files like llms.txt or publishing Markdown is not the same as traditional Google Search ranking advice.

Who should care most about AEO?

Developer platforms, SaaS companies, API providers, technical publishers, and any brand whose documentation or product content may be used by AI agents.

Conclusion

Osmani’s AEO guidance looks like an early but meaningful blueprint for the next phase of digital visibility. Content now has multiple consumers: people, search systems, answer engines, and increasingly, action-taking AI agents. Brands that adapt early will be better positioned to show up inside those workflows with clarity and trust. And if you are looking for a smarter way to build content for this shift, AIuthority is a strong place to start.