The secrets behind Google’s AI Optimization Guide

We've seen the daily AI updates: a new feature, a smarter AI model, a "hack" that will revolutionize your work, a token-saving "trick", a unique use-case you haven't thought of, etc. Our copywriters, UX/UI, motion and graphic designers, and web developers have all been absorbing AI updates every week. Naturally, Google's AI Optimization Guide was just another piece of information to absorb, but when we read it carefully, it revealed more than just steps to be cited by AI.

More answers, fewer clicks

Before getting into the guide and its secrets, let’s take a quick look at why AI Optimization is so important for Google, for us, and for you.

Because of AI, we can now search for something online and get the answers we need without clicking and being redirected to another site. This is called the zero-click economy: the clicks we once valued are being displaced by citation rates.

By July 2024, 60% of Google searches ended without a single click (Forbes). We are sure that as of today, that number is way higher. For Google specifically, 58% of all search results include an AI Overview (Forbes), which allows an easy zero-click search.

Before this, users used to find your site through page ranking position. The higher up you show, the more chances you’ll get noticed and clicked on. Now, AI is actively choosing what the user gets and where it is sourcing it from. This is the most important thing to do in the zero-click economy: optimizing your site to be chosen positively by AI.

AEO is the practice of optimizing your site so AI can find you, read you, trust you, and, finally, cite and recommend you, and that’s what Google’s guide was meant to cover.

Inside Google’s AI Optimization Guide

Most AEO experts recommend the same framework: structure your content so AI can extract it cleanly, build authority so AI trusts you, and write specifically so AI retrieves you for the right queries.

Google recommended exactly the opposite: don’t bother creating AI-readable sites, keep focusing on SEO, and write for people, not machines.

AEO experts were confused by Google’s stance. It contradicts what we've tested and the results we've seen. However, this is only Google’s recommendation for their own search engine. Google never states that this guide serves as a framework for other LLMs like ChatGPT and Claude.

Reading between Google’s lines

We started thinking about what this information would mean for our business. We consider AI’s preferences when we design sites and write content because we’re aware of how much visibility comes from it, so here are some points made by Google that made us stop and rethink our work:

Opinion over commodity

Google mentions that “creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search more than any of the other suggestions”. Up to this point, we agree that content must bring value to the reader. However, we also balance this with client briefs, content goals, and desired AI metrics.

Google suggests creating opinion-based content, using your unique experience as a source instead of what's already out there. 

This type of content is valuable, of course, and right in theory. However, in practice (especially in our practice as an agency), many client businesses are limited to factual content because of their category, regulatory environment, or audience. Industries like healthcare and finance are conservative and can’t create blog posts based on hot takes. Other industries might be more flexible, but most clients will push for strong sources as the foundation for content. Injecting a unique insight can add value, but it will mostly rely on a fact or shared experience.

AI “hacks” won’t work

In their “Mythbusting” section, Google mentions a list of AEO practices and notes that they “aren't effective or supported by how Google Search actually works.” The list suggests you don’t need AI-readable files or to structure and rewrite content to be cited and mentioned. 

Google calls LLMs.txt files “unnecessary”, but at least for other models like Perplexity and OpenAI (ChatGPT), we know LLMs.txt pages are present in their retrieval patterns. Some sources have even spotted Google Search Console crawling their LLMs.txt files. Wix, which implemented LLMs.txt across their platform, stated:

“Pages cannot be indexed without being crawled, meaning that these pages are not being ignored by Google. Those who have implemented LLMs.txt can observe activity on these pages in Google Search Console to see how and when they are being crawled. Let’s be clear, the intended audience for LLMs.txt is LLMs, not classic search engines, but the widespread misinformation that Google does not crawl LLMs.txt or that it is completely ignored is demonstrably false.” (Wix)

Whether or not Google officially endorses LLMs.txt, the evidence across other platforms is clear enough: the cost of implementing it is low, and the cost of being misread by an AI system is real.

Search rankings feed AI

Google also answers the question of SEO relevancy for generative AI search. They reveal that they use RAG (Retrieval-augmented generation) to improve the quality of AI responses. This technique consists of getting the relevant, up-to-date pages from their search ranking system and reviewing specific information from those retrieved pages to generate a response.

This fact makes us think that depth and specificity beat breadth. Site pages that answer specific questions could be more likely to be retrieved for specific queries, while broad pages compete with more sites for more questions. Google also mentions they generate sub-queries (query fan-outs) to fetch additional information, which also sustains our position of creating deeper and more specific content instead of several individual pages.

Javascript

Google's guide mentions that JavaScript works fine for AEO and even recommends following their JavaScript SEO best practices. Those practices exist for Googlebot, which has been rendering JavaScript for over a decade. LLMs like Claude, ChatGPT, and Perplexity that collectively represent the majority of AI traffic, have been documented as having trouble rendering JavaScript (clickrank). If your content loads dynamically, non-Google AI systems may never see it, so server-side rendering becomes a visibility consideration.

Can Google keep up with its own pace?

Google launched a new feature just eight days after publishing the guide. "Preferred Sources" is a visible badge users can give to the publishers (sites) they trust. These sites will get prioritized and surfaced more on their AI answers, which means a higher citation rate. Some sites are now encouraging users to add them as a preferred and have simplified the process by adding a CTA button. This feature is expected to become a ranking signal inside AI answers, and Google's guide has not been updated with this information.

The takeaway

AI is moving at an unprecedented pace, and with so much information available, we rely on our experience and what other agencies and experts are testing. After analyzing Google's guide, we drafted our own conclusions:

• For content writing, opinionated, first-person, experience-based content is stronger than ever, if the client's industry allows it. If not, specificity and depth are the next best lever.
• For UX and design, structure matters for retrieval. Pages that answer one thing well are better citation candidates than pages that cover a topic superficially. Page architecture should stay structured for clarity and hierarchy.
• For web dev, non-Google AI systems (other LLMs) can't render JavaScript correctly for now. If your content loads dynamically, ChatGPT, Claude, and Perplexity may never see it.
• Across all departments, AEO metrics like citation rate, mention rate, share of voice, and sentiment score are as important as traffic. A page that gets cited 50 times in AI answers and drives zero clicks is doing real work; we just need different tools to achieve it.