What Google Actually Published
The guide is at developers.google.com/search/docs/fundamentals/ai-optimization-guide. It is an official Google Search Central document, which makes it authoritative in a way that blog posts and consultancy pitches are not. Three things it does:
1. Confirms that Google's generative AI features (AI Overviews, AI Mode) ride on the same core Search index that classic Google Search uses. The underlying ranking systems are the same systems. 2. Lays out the techniques those AI features use to find and surface content: retrieval-augmented generation (RAG, also called grounding) and query fan-out. 3. Explicitly lists the AEO/GEO tactics that DO NOT work and that you can stop doing.
The combination matters. Google is saying: classic SEO IS GEO. There is no separate AI-specific optimization surface to chase. Your existing SEO foundation is the lever for AI Overview visibility, not a thing you add on top of it.
How AI Overviews Actually Pick Pages (in Plain English)
Two mechanisms do the work, both grounded in Google's core Search index:
**1. Retrieval-Augmented Generation (RAG, or grounding).** When a user asks something, Google's existing Search ranking system retrieves the most relevant pages. The AI model then reads those pages and generates an answer grounded in their content, with clickable citations back to the source pages. Your page being one of the cited pages is what AI Overview visibility actually means.
**2. Query fan-out.** The model also generates concurrent related queries to fetch more context. If a user asks "how to fix a lawn that's full of weeds," the fan-out queries might include "best herbicides for lawns," "remove weeds without chemicals," and "how to prevent weeds in lawn." Each of those triggers a fresh retrieval against the same Search index. A page that ranks well across the cluster of related queries gets more chances to be cited.
The practical implication: ranking in classic Google Search for the cluster of buyer queries IS the lever for AI Overview visibility. The two are not separate games.
Five Things You Should Do (Per Google)
The guide lists best practices for AI optimization. They are the same best practices for classic SEO, restated:
**1. Create unique, non-commodity content.** This is the single largest factor. Google's contrast example: "7 Tips for First-Time Homebuyers" loses to "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line." Commodity content built from common knowledge is replaceable. First-hand experience with specific detail is not. Sentie's content marketing strategy explicitly weaves first-hand operational evidence through every piece for exactly this reason.
**2. Build and maintain a clear technical structure.** Crawlable, indexable, meets technical requirements, follows JavaScript SEO best practices if you use a JS framework, has good Core Web Vitals. None of this is new; all of it still matters.
**3. Optimize local business and ecommerce details.** Where relevant, Merchant Center feeds and Google Business Profile let your products and locations show up in AI responses. For local-service businesses (HVAC contractors, dental offices, law firms), this is the under-leveraged lever.
**4. Use semantic HTML where it helps humans.** Not required for perfect parsing; useful for screen readers and increasingly for browser agents that interact with your site. The standard you should aim at is readability, not perfection.
**5. Organize content so readers can actually navigate it.** Paragraphs, sections, headings, structure. The thing that helps humans read your page is also the thing that helps AI systems extract from it.
Notice what is missing from this list. No mention of llms.txt. No mention of chunking. No mention of AI-specific page variants. Those are not on Google's recommended-actions list because they do not work.
Five Things Google Says You Don't Need to Do
This is the part of the guide that will surprise anyone who has been on the receiving end of AEO/GEO sales pitches:
**1. llms.txt files and "special" markup.** Quoting Google: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search." If a consultancy is selling you on building an llms.txt file as a ranking lever, they are selling you a tactic Google specifically says does not matter. Sentie deleted our llms.txt on Day 1 of our SEO campaign for this exact reason.
**2. Chunking content into tiny pieces.** Google: "Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users." There is no ideal page length. Write the length the topic needs. The chunking advice is from an older era when search engines did less context-aware retrieval.
**3. Rewriting content just for AI systems.** Google: AI systems understand synonyms and the meaning behind what people are searching, not just exact keyword matches. You don't have to create a separate page for every long-tail variant of how someone might ask. Writing for humans is writing for AI search.
**4. Seeking inauthentic mentions.** Google's spam systems explicitly block manufactured citations. Buying mentions or building inauthentic backlink networks is a strategy that worked in 2014 and gets you penalized in 2026.
**5. Overfocusing on structured data.** Schema.org is useful for Google's rich results (those expandable answer boxes), but Google explicitly says it is NOT required for GEO. Keep schema where it helps rich results. Stop treating it as a magic AI Overview lever.
What This Means for Your Business
If you have been paying a consultancy for AEO/GEO services, audit what they actually shipped. If the deliverables were "we built your llms.txt, set up chunked content, added AI-specific markup, generated long-tail keyword variant pages," you paid for tactics Google specifically says do not work.
If you have been DIY-ing your SEO and worrying that you are missing the AI-specific layer, you can stop worrying. The lever for AI Overview visibility is the lever for classic Google Search visibility, and it is the same lever you have been pulling. Pull it harder.
The practical priorities for the next 90 days, in order:
**1. Audit your content for commodity vs. non-commodity.** Walk through your top 20 pages and ask: does this page have first-hand insight no one else has? If not, rewrite or kill it. Generic listicles are the highest-risk format right now because AI systems can generate equivalents trivially.
**2. Sharpen the topical clusters you actually own.** Google's Web Guide test (publicly visible in May 2026) increasingly groups results by inferred subtopic, meaning cluster authority compounds. Build 5-10 pages around a tight cluster of buyer queries rather than 50 pages spread thinly across topics.
**3. Strengthen technical fundamentals.** Crawlability, Core Web Vitals, JavaScript SEO, mobile responsiveness. The unglamorous work that still moves the needle.
**4. Get your local business details current.** Google Business Profile, NAP consistency, accurate hours. AI Overviews surface this data in local-intent queries; out-of-date data costs you those citations.
**5. Watch the agentic experiences section.** Browser agents and emerging protocols (Universal Commerce Protocol, for example) are how AI search is expanding past text. Be ready, but do not contort your site for them yet.
What Sentie Is Doing Differently After Reading This
We rebuilt Sentie's SEO strategy explicitly against this Google doctrine on the day it was published. Specifically:
**Confirmed correct decisions** (already shipped): - Deleted llms.txt on Day 1 of our SEO campaign - Built our 100+ industry and solution pages around first-hand operational specifics rather than generic listicles - Pruned 130 thin pages from our sitemap to concentrate crawl budget on substantive ones - Wove first-hand dogfooding evidence into every blog post
**Doubled down on** (continuing investments): - Cluster cornerstone pieces (the "trades" cluster, the "professional services" cluster, the "commerce" cluster). See [/blog/what-is-a-custom-ai-agent](/blog/what-is-a-custom-ai-agent) and [/blog/what-is-a-business-brain](/blog/what-is-a-business-brain) for examples. - Technical fundamentals via daily monitoring of Search Console, Core Web Vitals, and indexation - Schema markup where it helps rich results but no longer treated as the AI Overview lever
**Explicitly stopped**: - Treating AEO/GEO consultancy claims as authoritative. Google Search Central is the only authoritative source. - Investing in any tactic not on Google's recommended-actions list. - Building separate "AI-optimized" page variants. The classic version IS the AI-optimized version.
The full strategy doc and decision log lives in our internal vault and gets refreshed every time Google publishes a new authoritative source.
The One-Paragraph Takeaway
Classic Google SEO is GEO. RAG and query fan-out ride on the same Search index. llms.txt, chunking, AI-specific markup, and most AEO consultancy tactics do not work. The lever is unique non-commodity content plus clear technical structure plus cluster authority plus accurate local/product data. The same things that made you rank well in 2018 are what get you cited in AI Overviews in 2026. The difference is that first-hand specificity matters more, commodity content is more replaceable, and the cluster structure compounds harder. Build the right kind of site and the AI search engines will find it. Do not let anyone sell you the hacks.