Google Just Leaked the SEO Roadmap Agencies Charge Thousands For

For two years, an entire industry sprang up around a new acronym: GEO, or generative engine optimization, the supposed art of ranking inside AI answers like ChatGPT, Gemini, and Google's AI Overviews. Agencies started charging thousands a month for it. Then Google published their own official guide on optimizing for generative AI search, and most of what the GEO industry sells is listed in it as a myth. This lesson pulls the signal out of Google's document, then shows you the free Claude skill that does the real work. Sourced from Google Search Central's official generative AI optimization guide, updated 2026-05-15.

What You'll Learn

• The five GEO hacks Google officially called myths, the ones your agency is still billing you for.
• The only two levers Google says actually rank you in both Search and AI answers.
• How to install and use Claude SEO, the free open-source skill built directly around Google's guide.
• A 30-day execution plan and a pre-flight checklist so the guide does not just stay a guide.

The one-line takeaway: in Google's own words, optimizing for generative AI search is optimizing for the search experience, and thus still SEO. There is no separate GEO game. There is good content on a clean site, and there is everything people are trying to sell you.

What to Ignore (Google's Own Myth List)

Google dedicated a full section of the guide to debunking the most common AEO and GEO tactics. These are the things you can stop paying for today.

• Myth 1, llms.txt and other special AI files: You do not need to create new machine-readable files, AI text files, or special markup to appear in generative AI search. Google may crawl many file types, but that does not mean any of them get special treatment. The llms.txt file does nothing for your Google visibility.
• Myth 2, chunking your content: There is no requirement to break your content into tiny pieces for AI to understand it. Google's systems already understand multiple topics on a single page and surface the relevant part. There is no ideal page length. Write for your reader, not for a chunking theory.
• Myth 3, rewriting your content just for AI: You do not need to write in some special way just for generative AI search. Google's systems understand synonyms and meaning, so they can connect people to your content even when it does not use the exact words of the query. Stop reformatting good pages into robotic, AI-bait versions of themselves.
• Myth 4, obsessing over structured data and schema: Structured data is not required for generative AI search, and there is no special schema markup you need to add to show up. It is still worth using for rich results in regular Search, but it is not the AI-ranking lever the industry treats it as.
• Myth 5, chasing inauthentic mentions: Seeding fake mentions of your brand across blogs, forums, and videos is not the hack it looks like. Google's ranking systems prioritize high-quality content and actively block spam, and the AI features depend on both. Manufactured buzz gets filtered, not rewarded.

Do the math: if your SEO or GEO retainer is built around llms.txt files, content chunking, AI-rewrites, schema packages, or mention-building campaigns, you are paying thousands a month for things Google just told you do nothing.

What Actually Works (the Two Real Levers)

After clearing the myths, Google's guide is almost boringly simple about what does matter. It comes down to two things.

• Lever 1, unique, non-commodity, people-first content: Google says this single factor will influence your presence in generative AI search more than anything else in the guide. The key word is non-commodity. Commodity content (think "7 Tips for First-Time Homebuyers") restates common knowledge anyone or any AI could produce. Non-commodity content brings a first-hand point of view, real experience, and insight that goes beyond the ordinary. If a generative model could have written it without you, it will not earn you the citation.
• Lever 2, a clean, crawlable technical structure: AI answers are built on Google's core Search index using retrieval-augmented generation, so a page can only be cited if it is indexed and eligible to appear with a snippet. That means the fundamentals still rule: be crawlable, follow JavaScript SEO basics if you use frameworks, deliver a good page experience across devices, and reduce duplicate content. Boring, but it is the entire technical half of the game.

The test Google gives you: ask one question of any page. Is this content my visitors would find genuinely satisfying? If yes, you are aligned with the systems. If you are writing it to game an AI, you are on the wrong track.

How AI Search Actually Pulls Your Content

Two mechanics from Google's guide explain why the two levers above are the whole game.

• Retrieval-augmented generation (RAG), also called grounding: The AI relies on Google's core ranking systems to retrieve relevant, up-to-date pages from the index, then generates an answer from them with clickable links back to the sources. No index presence, no citation.
• Query fan-out: The model spins up a set of related sub-queries to gather more context. Ask it how to fix a weedy lawn and it also fetches results for best herbicides, removing weeds without chemicals, and preventing weeds. Broad, genuinely useful coverage of a topic wins more of these than a single thin page.

Why ChatGPT is the same game: when you ask ChatGPT a question, it runs a web search and pulls from what it finds, the same loop. Win the underlying search and you win the AI answer, on every engine.

The Operator Move: Claude SEO

Knowing the two levers is one thing. Executing them across a real site is what agencies bill for. Instead of paying anyone, run your own sites through one free open-source Claude skill that does the technical audit and the content-gap analysis, then hands you the plan. It is called Claude SEO, built by Daniel Agrici.

Claude SEO is a free, open-source SEO analysis plugin for Claude Code. It runs 25 sub-skills and 18 specialist agents in parallel across technical SEO, content quality (E-E-A-T), Schema.org, AI search optimization (GEO), local SEO, e-commerce, and international SEO. Every audit produces a prioritized action plan with falsifiable recommendations. It is built directly around Google's AI Optimization Guide, so it targets the exact two levers that matter and ignores the myths. It is MIT licensed, ships with 271 tests, and is completely free. Repository: github.com/AgriciDaniel/claude-seo.

How to install it (one time). Fastest path on Claude Code 1.0.33 or newer is the plugin install. On an older version, use the manual install instead.

# Option A, plugin install (Claude Code 1.0.33+), fastest:
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo

# Option B, manual install (macOS / Linux):
git clone https://github.com/AgriciDaniel/claude-seo.git
bash claude-seo/install.sh

How to use it. Start Claude, then point a command at your site URL. Start with a full audit, then drill into whatever it flags.

# Start Claude Code
claude

# Full site audit, parallel agents return a prioritized action plan
/seo audit https://yoursite.com

# Deep single-page analysis (on-page, content quality, schema)
/seo page https://yoursite.com/your-page

# AI search optimization (passage citability per Google's guide)
/seo geo https://yoursite.com

# Content quality / E-E-A-T scoring
/seo content https://yoursite.com

# Technical SEO audit across 9 categories
/seo technical https://yoursite.com

Map the commands to the two levers. Lever 1 (content): run /seo content for the E-E-A-T gaps and /seo geo for AI-citability. Lever 2 (technical): run /seo technical and the full /seo audit. Ignore everything that does not roll up to those two. A full audit runs in about three minutes and costs nothing. Full credit to Daniel Agrici for building and open-sourcing Claude SEO.

Your 30-Day Execution Plan

• Week 1, run the audit and fix the technical floor: Run /seo audit on your site, then work the technical list it returns. Indexability, crawlability, page experience, duplicate content. This is the eligibility gate for every AI citation.
• Week 2, kill or upgrade commodity pages: Use /seo content to find the pages that just restate common knowledge. Either inject a genuine first-hand point of view and real experience, or consolidate them. Commodity content cannot win fan-out.
• Week 3, build one deep, non-commodity asset: Pick the topic you actually have lived expertise in and write the definitive, experience-led piece on it. This is the single highest-leverage move in Google's guide.
• Week 4, cover the topic broadly, not thinly: Run /seo geo to check passage-level citability, then map the real sub-questions a person asks around your topic and answer them genuinely, without spinning up junk pages just to chase keywords.

The Pre-Flight Checklist

• Cancel any llms.txt, AI-file, schema-package, or AI-rewrite line item that promised AI ranking.
• Install Claude SEO (github.com/AgriciDaniel/claude-seo) and run /seo audit on your site.
• Confirm every important page is indexed and eligible for a snippet, then fix crawlability, page experience, and duplicate content.
• Run /seo content to find commodity pages and either add first-hand experience or consolidate them.
• Publish one deep, non-commodity asset on a topic you have real expertise in.
• Run /seo geo and genuinely answer the fan-out sub-questions around that topic.
• Re-run /seo audit monthly and treat the readiness plan as your only roadmap.

The whole guide in one sentence: ignore the hacks, write things only you could write, keep your site clean and crawlable, and let one free Claude skill keep you honest.

Go Deeper

Grab the printable PDF version of this guide: https://drive.google.com/file/d/1SLE0hMRU_BXFs0wCL51YmoClh8OMPu99/view?usp=drivesdk

Want the whole system installed live, with Cooper on the call? Build Your Own AI, the two week sprint runs the first two weeks of August.

Or get a new working tool every Monday inside The AI Operator community on Whop.