For two years, "answer engine optimization" was practiced in the dark. Vendors sold theories. Nobody could check them.
On May 15, 2026, Google published a resource for optimizing for generative AI in Google Search, written for "website owners, SEOs, and developers." It's now the closest thing the field has to a primary source.
What the guide covers#
Per Google's announcement, the new guide includes:
- Guidance on providing valuable, unique, non-commodity content
- Tips on local, shopping, image, and video content
- Mythbusting common "AEO/GEO" misconceptions
- Initial guidance related to AI agents, described as "a quickly emerging and evolving space"
- Confirmation that SEO best practices remain relevant and foundational to success with generative AI features
That fourth bullet is the one to watch. Google is signaling that agentic access, meaning machines acting on behalf of buyers, is a live surface, and that its own guidance there is early rather than settled.
The correction nobody wants to hear#
The single most important phrase in Google's summary is non-commodity content.
The dominant AEO tactic of 2024 and 2025 was volume: generate hundreds of question-shaped pages, wrap them in FAQ markup, hope a model picks one up. Google's guidance points the other direction. Commodity content, meaning text that restates what a model already knows, has no reason to be cited. A model doesn't need your paraphrase of a public fact.
"Non-commodity means you know something the model can't infer: your pricing, your outcomes, your service reality. That's not a content strategy problem, it's whether your company writes down what it actually knows."
Andrew Bethel, COO, Perfectus Labs
What earns citation is information that exists nowhere else: your pricing, your process, your service area, your data, your results, your firsthand experience of the work. That's the same E-E-A-T logic Google has argued for years, now applied to a surface where the penalty for being generic isn't a lower rank but total omission.
What this validates#
If you have been doing the boring version of this work, the guide is good news:
- SEO fundamentals still compound. Crawlability, structure, speed, and clean information architecture remain prerequisites, not legacy.
- Local, shopping, image and video content is explicitly in scope. Answer surfaces are multimodal; text-only strategies leave coverage on the table.
- There's now a citable standard. You can settle internal arguments with a link instead of an opinion.
What this kills#
- Buying "GEO scores" from tools that can't see inside Google's systems
- Mass-produced question pages with no proprietary substance
- Any pitch that positions AEO as separate from, or a replacement for, SEO
How we read it at Perfectus Labs#
We built Cognos on a thesis this guide supports: the winner of an AI answer is the source that's easiest to verify and hardest to replace. So the work isn't "write more." It's:
- Make your unique facts machine-legible: prices, coverage, credentials, outcomes, named humans.
- Make them consistent everywhere a model can check them.
- Make the page that holds them faster and cleaner than any competitor's equivalent.
- Measure inclusion, not vanity rankings.
Read the guide yourself. Then ask a harder question about every page you own: if a model deleted this page, would the internet lose anything? If the answer is no, that page will never be cited, and no amount of markup will change it.
Source: Google Search Central, May 15 2026, posted by John Mueller.
