Google’s Official Guide to Generative AI Search: The Most Interesting Part Is What’s Missing

For over two years, the SEO industry tried to guess and reverse-engineer how AI Overviews and AI Mode work from the inside. Tests, patents, hundreds of articles and analyses. On May 15, 2026, Google published the answer. And the most interesting part is what’s not in it.

What happened and why Google’s Gen AI optimization guide matters

On May 15, 2026, a new page appeared in the official documentation on Google Search Central: “Optimizing your website for generative AI features on Google Search“. This isn’t a blog post with loose tips – it’s official documentation, in the same tier as the SEO Starter Guide, Technical Requirements or Spam Policies.

That distinction matters. Google’s blog posts are closer to commentary. Documentation is Google’s official layer of recommendations – more durable and operational than a single blog post, though of course it can still be updated.

The document targets site owners and anyone who wants to be visible in AI Overviews and AI Mode. And right at the start, it makes a claim that ends a certain industry debate. Or at least strongly reshapes it.

AEO and GEO are simply SEO. At least according to Google

A quote from the document:

Google says it directly: from Google Search’s perspective, optimizing for generative AI is SEO. No new discipline, no separate rulebook. Which practically begs for my favorite meme:

Does this close the debate? Not entirely – precision is due here. Google refers exclusively to its own systems: AI Overviews and AI Mode. This doesn’t cover ChatGPT, Perplexity or Claude – generative systems outside Google retrieve data differently, from different sources, by different criteria. There, GEO as a separate set of practices has stronger justification.

But in the context of Google – and Google remains the dominant content discovery platform – the term “GEO” applied to AI Overviews is now officially redundant. It’s SEO: the same thing as before, with new display surfaces.

In my book Marka osobista w czasach AI i generatywnego wyszukiwania (Onepress/Helion, June 2026 – in Polish) I put it this way:

Is GEO the new SEO? It’s more of an expansion of the playing field to include generative systems, which make decisions by criteria different from the classic ranking algorithm.

And Google has now sharpened that picture within its own ecosystem: generative search is not a discipline separate from SEO, but a new layer built on the foundations of classic search.

RAG and query fan-out. Google named the mechanisms out loud

In the new guide, Google describes in detail the two key mechanisms behind generative search:

RAG – Retrieval-Augmented Generation. Google defines RAG (using the term grounding interchangeably) as a technique in which the AI system retrieves current, relevant pages from the search index using the existing ranking systems, and then generates the answer on top of them. The document’s own words: “relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index”.

That’s an important sentence. It means RAG in Google Search relies on the search engine’s core ranking systems and the same index that classic results use.

• If your page is indexed, eligible to appear in Search, and answers the intent well – it can become a candidate for use in AI Overviews.
• If it isn’t indexed, or isn’t eligible to appear in search results – it drops out at the gate.

Visibility in classic search doesn’t guarantee a citation, but it raises the odds – because AI features draw on the same index and the same ranking systems.

I wrote more about how RAG picks its sources and what that means for your visibility in Your Brand in LLMs: What RAG Knows About You.

Query fan-out. Google describes it as a set of parallel, related queries the model generates to gather additional results. The document’s example: if a user asks “how to fix a lawn that’s full of weeds”, the system may generate sub-queries such as “best herbicides for lawns”, “remove weeds without chemicals” and “how to prevent weeds in lawn” in parallel.

I described this mechanism in my book in the context of Liz Reid, VP Head of Search at Google, who used the term query fan-out at Google I/O 2025:

AI Mode uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf.

Now query fan-out has entered the official documentation – which means Google treats it as a stable element of the architecture.

What does this mean in practice?

Your content doesn’t have to answer the user’s exact query word for word. The system decomposes the question into components and looks for an answer to each one separately. The deeper you cover a topic – from different angles, answering both the explicit questions and the ones the user never phrased – the more fan-out paths you can intersect.

In my assessment, this mechanism strengthens the case for topical authority: not through producing dozens of variants of the same content, but through real, deep topic coverage.

Five “hacks” Google just invalidated

The entire “Mythbusting generative AI search” section of the document is Google’s official answer to years of industry speculation. Google lists five things you don’t need to do. And in a few cases, the matter is less obvious than it looks at first glance.

llms.txt and “special markup”

Google writes plainly: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.” The llms.txt file – an idea that gained traction in 2024 as a way of “talking to AI” – carries no special meaning for Google Search.

Google adds a nuance: the search engine can discover, crawl and index various file types (including text files and Markdown), but that doesn’t mean it treats them in any special way.

In other words: llms.txt won’t hurt, but it gives no advantage in AI Overviews or AI Mode. If someone is creating it specifically for the Google ecosystem – that’s wasted time.

Content chunking

Google writes: “There’s no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users.”

And here a tension appears that I want to address head-on. In my book and in the BLUF article on this blog, I write about composing content in self-contained sections – about each section being understandable in isolation from the rest of the article, because RAG retrieves a fragment and evaluates it independently.

Does Google contradict that? No. Google is talking about something else: mechanically slicing content into micro-sections in the belief that AI can’t handle longer text. That’s “chunking” as an optimization technique – artificially short sections, content diced into miniature fragments, because someone decided “AI likes short pieces”.

Writing so that every section forms a self-contained answer to a question is simply good composition. Google itself writes, in the same document, that content should be “organized by paragraphs and sections, along with headings that provide a clear structure”. Section structure with informative headings is not an AI hack. It’s a standard of good writing that the algorithm simply processes better.

A reservation: this is Google’s position on its own systems. Other language models – ChatGPT, Perplexity, Claude – may process content differently, and there the discussion about optimal fragment structure will look different too.

Summing this thread up in one sentence: Google debunked mechanically dicing content into micro-fragments. It did not debunk writing so that every section carries standalone informational value.

Rewriting content for AI

Google writes: “You don’t need to write in a specific way just for generative AI search. AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words.”

This is another answer to the obsession with cramming in every long-tail phrase variant (and the strained attempts to blanket-cover query fan-out). Google says: the system understands synonyms and intent. You don’t need every query variant forced into the text. For experienced SEOs, that’s nothing new – synonym understanding has worked in Google for years. But an official confirmation in the generative AI context carries its own weight.

Again, though – writing content “for AI” is not the same as writing with citability in mind. Linguistic precision, sentences that stand on their own, BLUF – that isn’t “writing for AI”. That’s writing which is good for the reader and easy for the algorithm to process at the same time. The difference is subtle but real: Google debunked manipulating word choice, not deliberate text composition.

Inauthentic mentions

Google writes: “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem. Our core ranking systems focus on high-quality content while other systems block spam; our generative AI features depend on both.”

This is – at least officially – the closing of a topic that has circulated in SEO for years: mass-generating brand mentions on low-quality pages and articles. Google states directly that AI features rest both on the ranking systems (which assess content quality) and on the anti-spam systems.

Brand mentions still matter as a signal – but the real ones, the considered ones, earned in contexts the algorithm treats as credible.

Over-focusing on structured data

Google writes: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add. However, it’s a good idea to continue using it as part of your overall SEO strategy, as it helps with being eligible for rich results on Google Search.”

This sentence requires especially careful reading. Google is saying two things at once: there’s no special schema.org markup required to appear in AI Overviews – but structured data still helps with classic rich results and the overall SEO strategy.

E-commerce and local businesses – a separate section, separate tools

Google’s document isn’t limited to content and visibility in the classic sense. A separate section covers products, services and local businesses – and points to specific tools.

Merchant Center (including product feeds) and Google Business Profile remain the key data channels for AI. Google writes directly that generative answers can include product listings, product information and local business data – and that filling out these tools correctly raises your odds of appearing both in AI Overviews and in classic results.

A novelty in the document: Business Agent – a conversational experience on Google Search that lets customers talk to your brand directly in the search results. It’s a separate feature from the agentic experiences described later in this article.

For store owners and service businesses, the message is simple: product data and the business profile are not “extra” SEO elements. In the generative search ecosystem, they are the entry points AI builds shopping and local answers from.

Non-commodity content – one distinction that shifts the perspective

Before the details – one sentence from the document that deserves to ring out:

Read it again. Google says outright that unique, compelling and useful content will likely influence your visibility more than any other recommendation in the guide – more than technical structure, crawling or product data. That’s the strongest statement in the entire document. And it leads to a concrete distinction.

In the content-creation section, Google divides content into two types:

The difference doesn’t lie in writing quality, length or keyword optimization. It lies in whether a specific person with specific experience stands behind the content – and whether that shows in the article.

“7 tips for first-time homebuyers” can be written by AI, a copywriter or an intern, and the result will be similar – the content rests on common knowledge, and it’s easy to generate. “Why we waived the inspection and saved money: the story of our sewer line” requires a person who lived through it.

And though Google’s example is a surprising choice, it captures the point well: this is the content generative systems are looking for, because it gives them something they cannot generate from publicly available data.

Google also wrote: “Don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.” The sentence is phrased politely, but the message is brutal: if your article could have come out of a prompt – Google has no reason to cite it.

How does that translate into strategy? Anyone building visibility – personal brand, store, portal, service business – should ask themselves: does my content contain anything AI couldn’t generate on its own? First-hand experience, your own data, a point of view earned in practice – that’s the value the algorithm recognizes. I devote an entire chapter of my book to exactly this approach.

Images and video – an extra visibility surface

In the same context, Google adds an element that’s easy to miss: the search engine’s generative features can display images and video directly in AI answers. That means your site can appear in AI Overviews not only as a text link – but also as the source of a photo, an infographic or a recording.

Google points to its image SEO and video SEO best practices and states plainly: “If you’re already following our image SEO best practices and video SEO documentation, you’re already optimizing for generative AI search.”

In practice, this strengthens the argument for multimodal content. Your own photo from a project, an original infographic with data, a short implementation video – these aren’t aesthetic extras. They’re separate channels through which the algorithm can recall you.

What didn’t Google say? And why it’s the most interesting part of the document

Read the document once. Read it a second time. You’ll notice what’s not there:

• Not a word about E-E-A-T. The term that has been central to Google’s rhetoric on content quality since December 2022 – Experience, Expertise, Authoritativeness, Trustworthiness – doesn’t appear in the guide even once. Yes, the document links to the “creating helpful, reliable, people-first content” page, which is de facto the E-E-A-T page. But the term itself, the framework itself – absent. For a document meant to be the guide to generative search, that’s a telling omission.
• Not a word about entities. No reference to the Knowledge Graph, to author identity, to source verifiability. I wrote more about what a brand entity is and how to check yours in From String to Entity.
• Not a word about the author. In this particular guide, the question “who stands behind the content?” is never developed – though Google does link to the helpful, reliable, people-first content document, where authorship, E-E-A-T and trust are described directly. And that’s precisely why the omission is interesting: the AI Search guide shows the mechanics of accessing content, but doesn’t develop the source-reputation layer.

Google describes what the algorithm does (RAG retrieves pages, fan-out decomposes the query) and what the site owner should do (create valuable content, keep the technical layer healthy). But it doesn’t say how the algorithm decides whom to trust. The absence of a topic in one document doesn’t mean the topic doesn’t exist, though. Google simply chose what to talk about in this particular guide.

This is not an oversight. This is a deliberate choice.

The guide links to the helpful content page, yet the term E-E-A-T never falls. The document’s authors know the framework – they created it – so the omission is an editorial decision rather than a gap. And in my view it has a concrete reason: Google shows site owners the safe, official layer of recommendations: do good SEO, create unique content, keep the technicals in order, don’t play with hacks. As is its custom.

What it doesn’t give is an instruction like: “build an entity, and we’ll gladly trust you”. Hard to blame them – such a declaration would instantly turn into a checklist for mass manipulation. Something which, to be honest… we’re all trying to do anyway. Google communicates a clear direction; it doesn’t hand out instructions.

But for anyone who understands the mechanisms – and SEOs do – that silence says more than the text.

Google doesn’t need to spell that out in this guide. It’s enough to set it next to the helpful content documentation and the Search Quality Rater Guidelines to see that trust, authorship and source reputation remain part of the bigger quality picture in Google Search.

Agentic experiences – the direction Google is signaling

At the end of the document sits a section that looks like a curiosity at first glance, but is the announcement of a fundamental shift we’ve been expecting.

Google writes about agentic experiences – autonomous AI systems that carry out tasks on the user’s behalf, like browser agents analyzing screenshots, parsing the DOM structure, interpreting a page’s accessibility tree. Google points to its “agent-friendly website best practices“ guide.

And it mentions the Universal Commerce Protocol (UCP) – an open protocol meant to let AI agents complete transactions directly in the search results.

The guide gives it one sentence. But the context is wider: UCP is a protocol designed in collaboration with Shopify, Wayfair and Walmart among others, backed by more than 20 partners from payments and retail. The direction is unmistakable.

Why does this matter?

Until now, Google treated a website as a source of information – something the algorithm draws data from for its answers. Agentic experiences shift that relationship: the site becomes an interface an AI agent interacts with. It doesn’t read – it acts. It doesn’t cite – it completes a task.

For e-commerce, that’s a potential revolution in how a user gets from question to purchase. For service sites – an agent can book appointments or compare product specifications. For expert sites the impact is smaller for now, but the direction is clear: pages readable not only by humans but by AI agents will hold an advantage in the ecosystem Google is building.

So we’re no longer competing only for the click, not only for the citation – but also for being chosen by the agent.

That’s a topic for a separate article. I flag it here for one reason: the appearance of agentic experiences in Google’s official documentation is a clear message: get ready.

What follows from this document directly?

Google drew a map of generative search. It confirmed the mechanisms (RAG, query fan-out), knocked down a few industry myths (llms.txt, inauthentic mentions, chunking), said “do good SEO” – and, true to form, left the room.

The map Google leaves us shows the terrain: how the algorithm retrieves content, how it decomposes a query, what it counts as spam. But it doesn’t show the people – it doesn’t say who stands behind the content, why the algorithm cites one page and skips another, what makes a source credible.

And yet a large part of the game plays out exactly there. Not at the level of technicals (though those must be done well, as the document itself stresses), but at the level of who you are on the web – and whether the algorithm has grounds to trust you and recommend you.

Google’s guide tells you what to do. It doesn’t tell you who to be. And more and more often, that second question shapes your visibility.

One thing to keep in mind when reading Google’s documentation

Google published a guide, and it’s worth reading carefully. It’s also worth keeping some distance – not from this particular document, but from any documentation that tells you how to play a game whose rules its author sets.

The history of SEO is a history of scripts Google wrote – and rewrote. Every official recommendation was true at the moment of publication – and every one could change when Google’s priorities changed.

This guide is no exception. It’s solid, it’s useful, it’s worth implementing. But the best strategy is one that works regardless of whether Google publishes another document tomorrow that shifts the emphasis. Build a brand that has value not because the algorithm rewards it – but because people recognize it. Then every change of the rules plays in your favor.

Source: Google Search Central, “Optimizing your website for generative AI features on Google Search“, May 15, 2026.