Google Finally Published Its Official Guide to AI Search Optimization. Half of What People Are Doing Is Wrong.

google guide on optimizing website for AI Overviews

The guide kills off llms.txt, “chunking,” and AEO hacks. What it tells you to do instead is more interesting and harder.

Google published something last week that the SEO community has been waiting for since AI Overviews started eating into website traffic in 2024. Google’s actual position on how to optimize for AI search features so that your website might get quoted on AI overviews or suggestions.

The page is called “Optimizing your website for generative AI features on Google Search” and it’s sitting in Google Search Central’s documentation.

It covers AI Overviews, AI Mode, and the emerging world of agentic browsing. It’s the clearest Google has ever been about what works, what doesn’t, and unusually what you can actively stop worrying about.

This article breaks down what the guide actually says, cuts through the parts that sound like boilerplate, and adds the context that the official documentation can’t include which is the real traffic data, the publisher lawsuits, and the uncomfortable tension sitting at the center of all of this.

The context

In February 2026, Ahrefs published a study of 300,000 keywords and aggregated Search Console data going back to late 2023.

The finding is that AI Overviews correlate with a 58% reduction in click-through rates for top-ranking pages.

The position-one click-through rate for keywords that trigger AI Overviews dropped from 7.3% to 1.6%. For comparison, position-one CTR for keywords without AI Overviews dropped from 7.6% to 3.9% over the same period, a decline, but nothing close to the same magnitude.

Pew Research documented that users encountering an AI Overview click through to websites at an 8% rate, compared to 15% when no overview appears. That’s roughly half. Chartbeat, tracking more than 2,500 news sites globally, found that Google search referrals to those publishers fell 33% in 2025.

About 58% of all Google searches now end without a single click to any external website.

Penske Media, which owns Rolling Stone, Variety, and The Hollywood Reporter, filed an antitrust lawsuit.

The European Publishers Council lodged a formal complaint with the European Commission.

A third of publishers surveyed say they plan to block AI Overviews once Google provides the tools to do so.

On May 6, 2026, Google announced five updates designed to send more traffic back to publishers which include inline links, hover previews, “Further Exploration” article suggestions, subscription labels, and Community Perspectives from Reddit and forums. Those updates are, in most analysts’ readings, an indirect acknowledgment of the problem.

This is the environment in which Google published its optimization guide. Knowing that context changes how you read it.

What the guide actually says

Is SEO still relevant? Google says yes with a specific reason that matters

The guide opens by addressing the obvious question, and the answer is more technically specific than the usual reassurance.

Google’s AI features use something called Retrieval-Augmented Generation, or RAG. The model doesn’t generate answers from its training data alone.

It first pulls relevant pages from Google’s existing search index, then uses those pages to construct its response. The index that feeds AI Overviews and AI Mode is the same index that feeds traditional blue-link results. The ranking systems that determine which pages end up in that index are the same ranking systems that have always existed.

The practical consequence is that if your page can’t be found by traditional search, it also can’t be cited by AI features. Getting into the AI response requires being in the index first. The optimization pathway is the same even if the destination looks different.

Google also describes “query fan-out”, the AI automatically generates several related sub-queries to find supporting information for a complex question.

If someone asks “how to fix a lawn full of weeds,” the system might internally run concurrent searches for “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.” Your page doesn’t need to target the original query exactly. It needs to genuinely answer real variations of the underlying question.

What to actually do, the three things Google is serious about

1.Non-commodity content with a real point of view**

This is the most important section in the guide and the most honestly written one.

Google’s example for commodity content is “7 Tips for First-Time Homebuyers.” Generic advice, available from anyone, adds nothing that couldn’t be generated by a model.

Example for non-commodity content: “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line.” Personal experience, specific situation, insight you can only get from having done the thing.

The guide is specific about what a unique point of view actually requires. It means “create the content yourself based on what you know about the topic.” It means “first-hand review” rather than “summary of existing content.”

It explicitly says: “Don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.”

Google is describing a quality bar defined partly by being above what AI can produce. The implication is direct, content that an AI can generate is content that AI doesn’t need to cite. You need to bring something that a model can’t reconstruct.

What that looks like in practice: original data you collected, first-hand accounts of something you personally did, a counter-argument to the consensus view backed by your own experience, an analysis of something nobody else has analyzed. Not a synthesis of what others have already said.

2. Technical structure that makes your content crawlable

Google is clear that content that can’t be found and indexed can’t appear in AI features. This part of the guide doesn’t introduce anything new, it points to existing technical SEO documentation, but it does explicitly confirm that all of the standard crawlability requirements apply equally to AI feature inclusion.

A few points the guide makes explicit is that your page needs to be indexed and eligible to appear in search results with a snippet. Content blocked by robots.txt won’t be seen. JavaScript frameworks need to follow standard JavaScript SEO practices. Duplicate content wastes crawl budget and creates confusion. Page experience signals, load time, mobile rendering, navigability, still count.

One section that reads more forward-looking is “When it comes to semantic HTML, focus on human readability and don’t worry about perfect code.” The note about human readability also references screen readers in the same breath and Google is pointing toward the agentic browsing section that comes later.

3. Local and ecommerce details via Merchant Center and Google Business Profiles

If you run a physical business or an online store, AI Overviews and AI Mode can surface product listings, pricing, availability, and business details directly in the AI response. But they pull this information from Merchant Center feeds and Google Business Profiles, not from your website alone.

If your products are on your website but not in Merchant Center, they’re much less likely to appear in shopping-related AI responses. Same for business details. The guide points to Business Agent with a conversational experience that lets users chat with your brand through Google Search, as an emerging product worth knowing about.

What to stop doing

This is the part of the guide with the most direct practical value, because it lets you stop spending time on things that don’t work.

llms.txt files: Google doesn’t use them

There’s been a wave of enthusiasm in the SEO community around llms.txt, a text file placed in your website’s root directory that tells AI systems what your content is about, structured specifically for machine consumption. Vercel, Stripe, Shopify, GitHub, Anthropic, and OpenAI all publish them. The theory is that if AI systems read your llms.txt, they’ll understand your content better and cite it more.

Google’s position, now stated plainly that “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.”

John Mueller from Google had said informally for months that Google doesn’t use llms.txt for ranking. This guide makes it official documentation. Note what this says carefully, it says you don’t need llms.txt for Google’s AI features.

It doesn’t say these files are useless for other AI systems like ChatGPT, Claude, or Perplexity, which may handle them differently. If you’re optimizing across multiple AI platforms, not just Google, the calculus is different.

“Chunking” content is not a real optimization technique

“Chunking”, artificially breaking content into shorter, more digestible segments on the theory that AI systems prefer them, gets called out explicitly. Google says this is not effective and not supported by how their systems work.

The underlying concern the guide is addressing is real, that people have been restructuring their content based on theories about how AI retrieves information, creating pages optimized for machine consumption rather than human reading.

Google’s position is that this is a waste of time for its systems. Content organized for human readers, clear headings, well-written paragraphs, logical structure, is what both humans and Google’s AI want.

Creating content targeting “fan-out” queries: borders on spam policy violation

The guide acknowledges that people understand Google’s AI uses query fan-out, the automatic generation of related sub-queries, and some have responded by creating content specifically targeting each possible sub-query.

Google is direct: “focusing on other queries that people have asked, or fan-out queries” primarily to manipulate rankings violates the scaled content abuse spam policy.

Don’t build a page for every possible permutation of how someone might ask your core question. Build one page that answers the core question genuinely and well.

Seeking “inauthentic mentions” to boost AI visibility

Some consultants have been advising clients to get their brand mentioned on other websites specifically to influence AI training and retrieval. Google calls this out directly as something to ignore. Earned, authentic citations from genuine expert sources are worth pursuing. Manufacturing citations specifically for AI visibility is the same category of problem as manufacturing backlinks for traditional SEO.

Overloading on structured data

The guide doesn’t tell you to remove structured data. It tells you not to fixate on it as a primary AI optimization lever. Standard structured data practices, marking up articles, products, reviews, FAQs, how-to content where appropriate, remain useful. Adding mountains of custom schema hoping it gets parsed differently by AI systems is not a strategy.

The agentic browsing section: the part most people skipped

Google notes that browser agents, AI systems that navigate websites autonomously, like OpenAI’s Operator or Google’s own Project Mariner — access web content differently from traditional search crawlers. They don’t just read the page source. They analyze screenshots, inspect the DOM, and interpret the accessibility tree. They’re reading your website the way a sighted user does, not just parsing its HTML.

The practical consequence is that websites that work poorly for screen readers also work poorly for AI agents. Semantic HTML that clearly communicates the structure and meaning of your content helps agents parse it correctly. The guide explicitly links to web.dev’s guide on agent-friendly website best practices.

The more forward-looking piece is Google references the Universal Commerce Protocol (UCP) as an emerging standard that “will allow Search agents to do more.” UCP was announced earlier in 2026, co-developed with Shopify, with more than 20 companies endorsing it. Its purpose is to let AI agents complete commercial transactions on behalf of users like browsing products, comparing prices, and potentially purchasing directly through Google’s interfaces.

This is a hint about where the agentic search experience is heading. The optimization question isn’t just “how do I appear in AI Overviews?” It’s also “when an AI agent is browsing websites on a user’s behalf to complete a purchase, does my site give that agent what it needs?”

Structured product data, clear pricing, explicit shipping and return policies, and an accessible DOM start to look like table stakes for the agentic commerce future Google is pointing toward.

The tension Google can’t fully acknowledge in its own documentation

There’s something the guide navigates carefully.

Google’s official framing is that AI features are good for publishers because the clicks that do come through are higher quality, users who clicked through an AI Overview are more engaged, spend more time on-site, and convert better.

That’s true based on available data. Visitors arriving through AI Overviews do show lower bounce rates and longer time on-site in multiple studies.

But the conversion rate improvement doesn’t compensate for the volume decline in most publisher economics.

An 89% drop in click volume with a 30% improvement in session quality is still a very bad outcome for ad-supported publishers. A 58% average CTR decline industry-wide is not a quality story; it’s a structural revenue story.

Google’s optimization guide tells you how to compete for the citations and clicks that remain. It doesn’t address the underlying math that many publishers are currently living through.

For a solo content creator, a niche expert, or a brand with direct revenue models that don’t depend on raw traffic volume, the advice in the guide is genuinely useful. For a news publisher that monetizes page impressions through programmatic advertising at scale, “create better content” doesn’t solve the spreadsheet.

That’s not a reason to ignore the guide. It’s context for reading it accurately.

What this means for different types of sites

The guide’s advice lands differently depending on what you’re running.

If you’re a solo expert or niche specialist or a consultant, researcher, practitioner writing about what you actually do then Google’s AI optimization advice is almost perfectly aligned with your natural workflow.

Write about your real experiences. Share specific things you’ve learned from actually doing the work. Avoid recycling information that’s already widely available. This is the content that AI systems genuinely want to cite, and it’s also the content that builds a lasting audience regardless of search algorithm changes.

If you’re a small business or local service provider then make sure you’re in Merchant Center or have an up-to-date Google Business Profile. This is the most actionable single change many local businesses can make. The AI response for “best plumber in [city]” is increasingly pulling from Business Profiles, not from websites.

If you’re an ecommerce brand then your products need to be in Merchant Center feeds with accurate, structured data. Schema markup for products, reviews, pricing, and availability increases your chances of appearing in AI shopping responses. Keep your accessibility tree clean because browser agents increasingly walk ecommerce sites on behalf of users.

If you’re a content publisher then the guide’s advice is sound but insufficient for your situation. Getting cited in AI Overviews is worth pursuing. But the broader strategy shift to building direct audience relationships, diversifying traffic sources, investing in content that genuinely can’t be summarized in a paragraph, has to happen in parallel, not as a replacement for thinking about AI visibility.

The checklist from the guide

If you want to pull out the specific actions the guide recommends, here’s what it comes down to:

Publish content that draws on genuine expertise or first-hand experience, things a model can’t easily replicate.

Format it clearly with headings and logical paragraphs for human readers, not for machines. Make sure your site is indexed and shows a snippet in standard Google Search.

Ensure Googlebot can crawl everything that matters. Fix duplicate content. Meet Core Web Vitals for page experience. If you’re a local business, keep your Google Business Profile current. If you sell products, feed them into Merchant Center.

Stop building llms.txt files expecting Google to use them. Stop chunking content. Stop targeting fan-out query variations with separate pages. Stop manufacturing brand mentions for AI training purposes.

Start thinking about how AI agents browse your site. Make your semantic structure clear. Keep your accessibility attributes meaningful. Watch what UCP means for ecommerce as it rolls out.

The guide is at developers.google.com/search/docs/fundamentals/ai-optimization-guide. Read the original. This piece covers the key points, but the specific links to technical documentation in the crawling, JavaScript, and page experience sections are worth following if your site has any technical complexity.