Google recently released a set of guidelines for optimizing for generative AI (so-called Generative Engine Optimization or GEO). As always, Google has to balance providing guidance and preventing abuse with giving away too much information. To save you from reading between the lines, here are 5 key takeaways from my read:

Let’s start with the good news. Right at the beginning, Google confirms that:

The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems.

Naturally, you might have doubts, given how different AI Mode and the impact of our SEO efforts seem in 2026. It’s true that the SERP itself has changed and that LLMs operate differently from traditional information retrieval systems, but the key here is that Gemini-powered search — including FastSearch — still heavily relies on the search index (whether you call it “search grounding” or retrieval-augmented generation).

Is it SEO, GEO, or AEO?

According to Google, “... optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Ok, it’s a bit of a cheat, based on a shifting definition of search. Personally, I think “GEO” reflects some of the specific aspects of GenAI and LLMs, and can be a useful distinction. If that distinction helps you get the job done, then great. If it’s just another buzzword to pitch on LinkedIn, then good luck.

I’d love to stop the post at #1, but of course, it’s not that simple. Keywords and SERPs used to have a one-to-one relationship. Everything on that page was directly relevant to the keywords you typed into the box. Gemini-powered search uses query fan-out — your initial keywords are used to “fetch additional relevant search results,” and those are combined into an answer (whether it’s an AI Overview, AI Mode, or Web Guide page).

Think of query fan-out as a journey that builds on your initial query:

Someone who starts their journey with “What are the best wireless mice for gamers” might ask a follow-up question like “How long do wireless gaming mice last?” or want a human perspective, such as “Why do some gamers swear by wrist rests?”

Even in Web Guide results, you have to consider that result #7 on the page (for example) could be result #1 for a fan-out search. It’s still SEO, but any single SERP could represent half a dozen related searches condensed into one composite page.

In the mythbusting section, Google says: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” This isn’t a condemnation of structured data; it’s a reflection of one simple fact: search spiders and LLMs are hungry and need to ingest as much content and data as possible. The internet is messy, and Google can’t wait for webmasters to optimize their sites.

Structured data certainly doesn’t hurt, and it can have other benefits — including qualifying your site for rich results and feeding the Knowledge Graph. It’s also generally important to create well-structured, readable content for humans. Obsessively optimizing your markup, though, isn’t going to produce additional benefits.

We all know about E-E-A-T by now, and we’ve heard the “build great content” mantra for years. In just this one document, Google suggests you build content …

• That provides a “unique point of view”
• Is “helpful, reliable, and people-first”
• That “visitors would enjoy, find helpful, and feel satisfied with”

Sometimes, this kind of advice feels hopelessly vague — at best, it’s not tactical. There are two bits of advice, though, that stood out for me in this document.

Create “non-commodity” content
In other words, don’t write content that anyone could write with basic knowledge, and that’s likely to be lost in a sea of similar articles. This comes back to the two E’s in E-E-A-T: Expertise and Experience. When LLMs can easily summarize common knowledge, the one thing you can still provide is human perspective.

Serve the need, not the keywords
I’m paraphrasing a bit, but don’t let query fan-out go to your head and try to write content for every single way a human could ask a question or express a need. It’s not necessary, and it’s likely to devalue your site. In 2026, you also don’t need to “chunk” everything into micro-content to be easily digested by LLMs. Write for the broader topic, targeting key concepts and the user journey you’re trying to serve. More isn’t better.

The guidelines briefly mention “agentic experiences,” foreshadowing Google’s announcements at I/O 2026 (on May 19). What’s agentic SEO? — It’s SEO for agentic search. What’s agentic search? — It’s search enhanced by intelligent agents. What’s an agent? Well, it’s an intelligent assistant that does stuff for you.

Does that all sound a little vague? The challenge is that it is until it isn’t. In the coming year, we’ll likely see agentic experiences for performing specific tasks, like booking tickets or building custom dashboards. Until those materialize, though, “agentic SEO” isn’t a very helpful term. Think of it like rich SERP features. Some of them, like song lyrics, radically changed certain businesses overnight, but had no impact on 99%+ of searches.

This isn’t a very comforting thought, but it’s hard to prepare for agentic SEO in any general sense. We’re going to have to see which agents Google specifically releases, what impact they have, and, frankly, whether anyone wants them at all.

Sorry, that sounds cruel, but it’s the truth. Search-grounding means that much of the core SEO work you’re already doing will also help your appearance in GenAI/Gemini results. Generally speaking, you don’t have to optimize for GenAI, and over-optimizing on one dimension, like structured data, is going to be time poorly spent. That said, query fan-out and the divergence of 10 blue links into a rich mix of GenAI features, including direct AI Mode results — and, eventually, agentic search — mean that the classic SERP is now a much more complicated hybrid of results from multiple sources. We don’t have to panic, and we don’t have to forget everything we’ve learned, but we do have to adapt.