Google's AI Optimization Guide

Google's AI Optimization Guide: Clear words on AI search – with blind spots

Google has been silent for a long time. While the SEO industry has been getting tangled up for months in acronyms like GEO, AEO and LLMO, Mountain View has remained conspicuously tight-lipped on questions around AI Overviews and AI Mode. In mid-May 2026, the silence is broken. With the official Guide to Optimizing for Generative AI Features on Google Search the search giant is presenting a documented statement for the first time – and is thereby answering a question that has been preoccupying online retailers since the roll-out of AI Overviews: How does my website remain visible in a search where answers are increasingly generated directly?

A factual classification of the first official Google guide to AI search optimization

The guide is an important document. But it is also a typical Google document: What it says is clear, useful and, for the most part, reasonable. What it does not say is at least as revealing. Those who rely only on what Google officially says are optimizing for a slice of reality, not for reality.

In this article, we classify the new guide. We show what Google is now officially stating, what the company is clearing up, and where the guide has its limits.

What Google officially says: The core message of the guide

The guide is dated May 15, 2026 and is embedded in the English-language Google Search Central documentation. This gives the topic of AI optimization a permanent place alongside the classic SEO Starter Guide for the first time. That alone is remarkable. Google is signaling: Generative AI is no longer a special case, but part of search.

Google has apparently chosen to formulate the central message in a deliberately pointed way: SEO is not dead for AI search, but a prerequisite. According to Google, the AI features, above all AI Overviews and the new AI Mode, are based on the existing core ranking and quality systems. Those who are invisible in classic search will also not appear in generated answers.

What the Google AI guide essentially explains

In detail, the guide explains two technical mechanisms that are crucial for visibility in AI answers:

• Retrieval-Augmented Generation (RAG): The AI does not generate its answers from pure language-model knowledge, but bases them on current web pages determined by classic ranking. Google calls this “grounding”. In plain terms this means: The sources that perform well in the normal search index are also the ones that are cited in AI answers.

• Query fan-out: A single user query triggers several parallel search queries in the background. So if someone asks “remove weeds from lawn”, this additionally triggers searches for herbicides, mechanical methods and prevention. Visibility no longer results only from the one main query, but from the entire topic area.

The recommendations that Google derives from this sound familiar to any experienced SEO: unique, user-centered content with its own perspective, clean technical structure, indexable code, good page experience, sensible internal linking, high-quality images and videos. Anyone who takes the Search Essentials and the Helpful Content principle seriously is, according to Google, already optimizing for AI search.

Pointedly summarized from Google's perspective: GEO and AEO are not a new discipline. To a large extent, it is SEO under new auspices.

The mythbusting: What Google actively clears up

More exciting than the positive recommendations is the section in the guide that explicitly opposes certain optimization tactics. Google devalues several concepts that have been circulating in the GEO community for months:

• llms.txt files: According to Google, they are not required for visibility in generative Google search. A separate markup file for AI systems offers no advantage in the Google universe.

• Content chunking: Breaking content down into tiny sections for AI is unnecessary. Google's systems understand context across longer passages of text.

• AI-specific rewriting: Rewriting texts specifically for AI models, for instance by including every possible question phrasing in the text, is, according to Google, unnecessary and, if overdone, can even be classified as spam, i.e. scaled content abuse.

• Inauthentic mentions: The attempt to get yourself mentioned in a targeted way in forums, blogs and Reddit threads is seen by Google as not very effective. The internal spam systems would filter out such signals.

• Structured data as an AI lever: Schema.org markup is not required to appear in AI Overviews. It remains useful for rich results in classic search – but it is not a secret AI recipe.

These clarifications are valuable. They clear up a series of miracle cures that are circulating in the industry and that we at arboro regularly have to explain and classify in client meetings. Anyone who has relied on llms.txt or chunking as a universal silver bullet should read the guide carefully.

What the guide does not say: Four blind spots

As useful as the guide is, it is not a neutral document. It is the official view of a commercial company on its own product. Anyone who takes it seriously should therefore also put it into context. Four points deserve special attention.

1. The guide deals exclusively with Google

This is evident from the title, but is often lost in the discussion: The guide refers exclusively to AI Overviews and AI Mode in Google Search. It says nothing about ChatGPT Search, Perplexity, Claude, Microsoft Copilot or Brave Leo. These systems work with their own crawlers, their own training data, their own evaluation systems and their own answer mechanisms.

Anyone who observes today where online shops get their conversions from can see: Traffic from AI sources beyond Google is measurably increasing. An optimization strategy that focuses only on Google leaves a growing portion of AI-driven visibility unused. This is precisely where Generative Engine Optimization, or GEO for short, comes in as an independent discipline, even if Google would prefer to subsume this term under “SEO”.

2. llms.txt is indeed relevant for other LLMs

Google's statement that llms.txt is unnecessary applies to Google. It does not apply universally. Other providers actively rely on standards that give website operators control over how their content may be used for AI systems, whether via llms.txt, ai.txt or extended robots.txt directives for GPTBot, ClaudeBot, PerplexityBot and others. Anyone who wants to control these crawlers cannot avoid the corresponding files. We have already published a detailed guide on this in the arboro blog.

3. Google statements have historically been revised

Anyone who has been involved in SEO for a few years knows the pattern: Google makes a clear public statement and later revises or differentiates it. After years, E-A-T became E-E-A-T with the additional “Experience”. The Helpful Content Update was silently integrated into the Core Updates in 2024. The statement that clicks and CTR are not a ranking signal clearly began to wobble in the US antitrust trial. The Google Search API leaks of May 2024 revealed internal modules whose existence had in some cases been publicly denied.

This is not meant to discredit Google across the board. The official guidelines are still the best available source. But: they are a snapshot, not a final verdict. What is considered unnecessary today may be recommended tomorrow, and vice versa. Anyone who bases a strategy solely on the current wording of a Google document is building on shifting ground.

4. The self-interest remains invisible

Google has an economic interest in keeping online retailers inside the Google ecosystem. Unsurprisingly, the guide recommends tools such as Merchant Center, Google Business Profile, and the new Business Agent. It mentions the growing importance of agentic experiences via protocols such as the Universal Commerce Protocol, or UCP for short. All of this makes sense. But it is also a proposal with a business model.

The statement “GEO is just SEO” relieves Google rhetorically in an elegant way: it retains interpretive authority over the discipline and prevents AI optimization from being discussed as an independent field beyond Google. From arboro’s point of view, this is an understandable position, but it’s not the whole story.

Why GEO remains an independent discipline nonetheless

We take Google’s position seriously. Much of what is in the guide overlaps with what we implement in our GEO projects: a strong content foundation, technical cleanliness, unique perspectives, good page experience. Anyone who works solidly here has done a large part of their homework.

But GEO is more than that. It’s about visibility in a fragmented answer market in which Google is just one of several relevant players:

• Mechanics differ: ChatGPT draws on Bing-based indexes and its own web crawlers. Perplexity works in a very source-driven way with clear citations. Claude is more restrained with live web sources in the consumer product, but gains visibility via training data. Each system has its own logic for which content is cited or paraphrased.

• Brand mentions become a currency: What Google dismisses as “inauthentic mentions” for its own search works very differently in the training data of other models. There, in the long run, it also matters how often and in which contexts a brand is mentioned on the web, and not just how it is linked.

• Conversational search intent: Users who talk to an AI formulate differently than on Google. They ask longer questions, follow up, compare. Content that serves this conversational intent is a separate competency, independent of schema markup or chunking.

• Agentic commerce on the rise: AI agents that conduct research, compare products, and even prepare orders on behalf of users need machine-readable, precise information. This is where it will be decided which shops will become recommendations in the coming years and which will not.

GEO is therefore not a marketing invention to justify a new line item on the invoice. It is an honest recognition that search and recommendation are being distributed. And that pure Google optimization no longer covers this distribution.

What online retailers should concretely do now

Anyone who has read the guide and adds our assessment can clearly draw the consequences for their own strategy. From our consulting practice, we recommend:

Consistently audit the SEO foundations

Anyone with technical deficiencies, thin content, or a shaky page-experience profile is not yet optimizing for AI – they are already lagging behind in classical rankings. This is where the greatest leverage-to-effort ratio lies.

Make your own perspective visible

Experience-, expert-, and first-user content is what generative systems cannot produce out of themselves. Originally created content with a clear stance is the only form that will be cited in the long term.

Deliberately control AI crawlers via robots.txt

The major AI systems respect the robots.txt protocol and can be specifically addressed there: GPTBot and OAI-SearchBot from OpenAI, PerplexityBot, ClaudeBot and anthropic-ai crawl separately from the classic Googlebot. Anyone who wants visibility in these systems should not accidentally block access, but consciously decide which content is released for AI training and AI answers.

Build brand mentions strategically

Not artificially – that doesn’t work with Google anyway. But editorially, with trade publications, industry studies, guest contributions, and clean digital PR. These mentions are the foundation on which AI systems build trust.

Continue using structured data

Even if the guide does not make them mandatory for AI: they remain valuable for rich results in classic search, for Merchant Center, and, in the long term, for agentic commerce protocols.

Monitor AI Overviews and alternative answer systems regularly

Visibility in generated answers has to be measured separately – it does not appear in every standard Search Console view. Tools and routines are needed here to supplement click and impression data.

Factor in agentic experience

Anyone working in e-commerce should maintain product data, availability, and service conditions in a structured way so that AI agents can understand them – the investment will pay off over the next 24 months.

Conclusion: Take the guide seriously, but don’t mistake it for the whole truth

Google’s first official guide to AI search optimization is a good, clear, and in many respects overdue document. It debunks some popular myths and puts SEO fundamentals back at the center of the discussion, where they belong.

Decision-makers for the online shop should still keep three things in mind

First: The guide is an official position, not the complete truth. Google has revised statements in the past. That will also happen here. The practices mentioned in the guide are a good orientation today, not a guarantee for the next five years.

Second: Anyone who thinks about visibility only for Google is thinking too small. ChatGPT, Claude, Perplexity, and upcoming systems play by their own rules. An AI strategy that only addresses AI Overviews optimizes for a market share, not for AI search as a whole.

Third: The most valuable asset remains human substance: your own experience, your own position, your own data points. This is the form of content that does not emerge from nothing in any AI model. Those who invest here are better protected against any future algorithm revision than any technical optimization could ever provide.

arboro has been supporting online retailers for years in the transition from classic SEO to generative search. As one of the first providers in the DACH region with dedicated GEO services, we help you build visibility beyond Google, not only in Google’s AI Overviews, but also in ChatGPT, Claude, and Perplexity. If you want to know where your shop currently stands in AI search and where the biggest levers are: talk to us.

Sources

1. Google Search Central: Optimizing your website for generative AI features on Google Search, published on 15 May 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide

2. Google Search Central: Creating helpful, reliable, people-first content. https://developers.google.com/search/docs/fundamentals/creating-helpful-content

3. Google Search Central: Guidance on using generative AI content. https://developers.google.com/search/docs/fundamentals/using-gen-ai-content

4. Google Search Central: Search Essentials. https://developers.google.com/search/docs/essentials

5. arboro blog: robots.txt for Shopware 5 & 6 in the AI era. https://www.arboro.de/blog/robots-txt-fuer-shopware-5-6-im-ki-zeitalter-produkte-sichtbar-machen-crawler-kontrollieren-chancen-nutzen

6. arboro blog: Universal Commerce Protocol (UCP) and Agentic Commerce. https://www.arboro.de/blog/universal-commerce-protocol--und-agentic-commerce