GEO Is Mostly Snake Oil — Here's What Actually Works According to Google

There's a new three-letter acronym making the rounds at marketing conferences, in agency pitch decks, and across LinkedIn thought-leader posts: GEO. Generative Engine Optimization. Sometimes it goes by AEO — AI Engine Optimization. The pitch is compelling: AI is changing search, and you need a completely new optimization strategy or your brand will disappear from AI-generated answers.

Agencies are packaging GEO services at premium rates. Consultants are selling workshops. SaaS tools are launching with GEO-specific dashboards. And the tactics they're pushing — llms.txt files, "content chunking," AI-specific rewriting, mention farming — sound technical enough to be believable.

There's just one problem. Google, the company that built the largest AI search experience on the planet, has published detailed documentation on how to optimize for their AI features. And that documentation contradicts most of what the GEO industry is selling.

Let's walk through what Google actually says, line by line.

The GEO Sales Pitch

First, let's be fair about what's being sold. The GEO industry has coalesced around a set of tactics that supposedly help your content appear in AI-generated answers. The typical package includes:

• llms.txt files — A robots.txt-style file that supposedly tells AI crawlers how to read your site
• Content chunking — Restructuring your pages into discrete, AI-digestible blocks
• AI-specific rewriting — Reformatting content so AI models can "understand" it better (often meaning adding explicit Q&A structures, simplified language, or "AI-friendly" formatting)
• Mention farming — Getting your brand mentioned across forums, directories, and third-party sites so AI models "learn" about you
• Structured data optimization — Adding schema markup specifically to appear in AI features
• Long-tail query targeting — Creating pages for every possible conversational query variant an AI might process

Some of these ideas aren't crazy on their surface. Structured data is genuinely useful. Good content structure matters. But the framing — that these are new tactics for a new channel — is where things fall apart.

An important distinction before we go further: The problem isn't caring about AI visibility. Brands absolutely should understand what AI platforms say about them — that's a legitimate business need. The problem is the specific tactics being sold under the GEO label. Monitoring your AI presence, understanding where you're visible and where you're not, tracking what AI says about your competitors — that's intelligence. Creating llms.txt files and farming mentions — that's snake oil. Don't confuse the two.

Google Doesn't Use the Terms "GEO" or "AEO"

This is worth pausing on. Google has published an entire guide on optimizing for AI features in search. The title? "Optimizing for generative AI features on Google Search."

Not GEO. Not AEO. Not "Generative Engine Optimization."

These terms were invented by the marketing industry, not by the platforms themselves. That doesn't automatically make them wrong, but it should make you skeptical when someone positions GEO as a distinct discipline with its own best practices. Google doesn't recognize it as one.

More importantly, Google's guide opens with a statement that should give every GEO vendor pause: optimizing for AI features is fundamentally the same as optimizing for traditional search, because AI features use retrieval-augmented generation on the same ranking systems.

Read that again. The same ranking systems. Not a separate AI index. Not a different algorithm. The same one.

Google Debunks the Core GEO Tactics

Let's get specific. Google's AI optimization guide directly addresses — and dismisses — several of the most popular GEO tactics being sold today.

No Special llms.txt Files

There is no Google-recognized llms.txt standard. Google's crawlers (Googlebot, Google-Extended) follow robots.txt and standard web protocols. The idea that you need a special file to instruct AI models how to consume your content is not supported by anything in Google's documentation.

If a vendor is charging you to create and maintain an llms.txt file for Google visibility, they're charging you for something Google doesn't read.

No "Content Chunking" for AI Consumption

The notion that you need to restructure your content into discrete chunks specifically for AI consumption isn't backed by Google's guidance. Google's AI features pull from the existing index. The content that ranks well in traditional search is the content AI features will draw from.

Good content structure — clear headings, logical flow, answering questions directly — has been an SEO best practice for over a decade. Repackaging it as "AI content chunking" and charging a premium for it is, at best, renaming existing work.

No AI-Specific Rewriting Styles

Google's documentation doesn't recommend any special writing format for AI features. You don't need to write in a particular way so that AI "understands" your content better. Modern language models can parse natural human writing — that's literally what they were built to do.

The guidance is the same as it's always been: write clearly, cover the topic comprehensively, use natural language. If your content needs to be rewritten for AI to understand it, it probably needed to be rewritten for humans too.

No Inauthentic Mention Campaigns

Mention farming — the practice of seeding your brand name across forums, directories, and low-quality third-party sites — is not just ineffective for AI features, it's potentially dangerous. Google's spam policies explicitly address inauthentic engagement, and the Quality Rater Guidelines instruct raters to evaluate the reputation of a site based on independent, credible sources.

Fake mentions aren't credible sources. They're noise. And if Google's systems identify a pattern of manufactured mentions, you're not optimizing — you're spamming.

Structured Data: Helpful, Not Required

Here's where it gets nuanced. Structured data (schema markup) is genuinely useful. Google recommends it. It helps search engines understand the entities and relationships on your page.

But — and this is the part GEO vendors leave out — structured data is not required for your content to appear in AI features. Google's documentation is clear: AI features can and do pull from pages without any structured data. Schema markup is one signal among many, not a magic ticket into AI Overviews.

If a vendor tells you that adding structured data is the key to appearing in AI-generated answers, they're overstating its importance. It's a good practice. It's not a GEO strategy.

Don't Target Excessive Long-Tail Variations

One of the more damaging GEO tactics is the idea that you should create separate pages for every possible conversational query an AI user might type. "What's the best CRM for small business?" gets one page. "Best CRM small business 2026" gets another. "Which CRM should a small business use?" gets a third.

Google's documentation specifically warns against this. Their AI features use query fan-out — the AI generates related sub-queries to build comprehensive answers. It doesn't need you to have a page for each variation. One genuinely comprehensive page on a topic will serve multiple query variants.

Creating separate pages for each variation isn't optimization. It's doorway abuse, and Google's spam policies have a section dedicated to it.

How Google's AI Features Actually Work

Understanding why these tactics don't work requires understanding the architecture behind Google's AI search features. Google has been unusually transparent about this.

Retrieval-Augmented Generation (RAG)

Google's AI features — AI Overviews, AI Mode, and others — use a technique called Retrieval-Augmented Generation. Here's what that means in practice:

1. A user asks a question
2. The AI generates related sub-queries (query fan-out)
3. These sub-queries are run against Google's existing search index using the same ranking systems that power traditional search results
4. The top-ranking results are retrieved and provided to the AI model as context
5. The AI model generates an answer based on that retrieved context, with citations

The critical insight is step 3. The AI doesn't have its own separate index. It doesn't have special ranking criteria. It uses the same systems that determine whether you rank #1 or #50 in traditional search results.

The Same Ranking Systems Power Both

Google's ranking systems — RankBrain (query understanding), BERT and MUM (natural language processing), Neural Matching (concept matching), PageRank (link authority), and dozens of others — are the foundation for both traditional results and AI features.

This means the competitive moat for AI visibility is the same competitive moat you've been building (or should have been building) for years: domain authority, content quality, topical relevance, user experience, and trust signals.

There's no shortcut. There's no new channel to game before competitors catch on. If you rank well in traditional search for a topic, Google's AI features will likely draw from your content. If you don't rank well, no amount of llms.txt files or content chunking will change that.

What Actually Works — Straight From Google

So if GEO tactics are largely repackaged SEO (or outright ineffective), what does Google actually recommend? Their documentation is refreshingly straightforward.

Create Valuable, Non-Commodity Content

Google's guidance emphasizes content that provides genuine value — meaning content that a reader couldn't easily find elsewhere. This includes:

• Original research and data that only your organization can produce
• Unique perspectives from practitioners with real experience
• First-hand reporting that adds new information to a topic
• Proprietary analysis that synthesizes information in novel ways

Notice what's absent: there's no mention of writing for AI models, formatting for machine consumption, or using any specific structural template. The emphasis is entirely on the substance of what you publish.

If your agency client runs an e-commerce platform and has transaction data showing purchase trends, publishing that original data is more valuable than rewriting a generic "Top 10 E-commerce Trends" post in an "AI-friendly" format.

Build E-E-A-T (and Understand That Trust Is Central)

Google's E-E-A-T framework — Experience, Expertise, Authoritativeness, Trustworthiness — isn't new, but it's more relevant than ever for AI features. Here's why: when an AI model generates an answer and needs to choose which sources to cite, it's drawing from results that already passed through ranking systems that weigh E-E-A-T signals heavily.

Google's Quality Rater Guidelines make Trust the central element. A site can have expertise and authority, but if it's not trustworthy, its quality rating drops. Trust is built through:

• Accurate, well-sourced content
• Clear authorship and organizational transparency
• Consistent editorial standards
• Secure site infrastructure
• Honest business practices

For YMYL (Your Money or Your Life) topics — health, finance, safety, legal — the trust bar is even higher. AI features are especially cautious about citing sources for these topics, making E-E-A-T signals even more critical.

Comprehensive Topic Coverage With Natural Language

Rather than creating dozens of thin pages targeting individual query variations, Google recommends comprehensive coverage of a topic in natural language. One authoritative, thorough page on "CRM for small business" will outperform twenty thin pages targeting slightly different phrasings of the same question.

This aligns with how query fan-out works: the AI generates sub-queries and looks for content that addresses multiple angles of a topic. A comprehensive page naturally answers many of those sub-queries. Twenty thin pages each answer one, poorly.

Technical SEO Fundamentals

None of this is new, but it still matters:

• Mobile-first design — Google predominantly uses mobile-first indexing
• Fast load times — Core Web Vitals remain ranking signals
• Crawlable architecture — If Googlebot can't access your content, AI features can't use it
• Clean URL structures — Logical site architecture helps both crawlers and users
• HTTPS — Basic security is a trust signal

These aren't GEO tactics. They're the same technical SEO practices that have mattered for years. But they're often neglected by agencies that are too busy selling shiny new GEO packages to maintain the fundamentals.

Independent Reputation Signals

Google's systems evaluate your site's reputation based on independent sources: news coverage, Wikipedia mentions, expert reviews, industry awards, Better Business Bureau ratings, and similar signals.

This is the opposite of mention farming. Google isn't looking for your brand to appear on fifty random directories. It's looking for credible, independent sources that validate your expertise and trustworthiness.

You can't manufacture this. You earn it by doing good work, getting covered by real publications, earning genuine reviews, and building a legitimate reputation in your space.

The Spam Risk Nobody's Talking About

Here's where GEO tactics cross from "ineffective" to "actively dangerous."

Scaled Content Abuse

Google's spam policies identify scaled content abuse as a top concern. The definition: using automation (including AI) to generate large volumes of content primarily to manipulate search rankings, without adding substantial value.

Several GEO tactics walk right up to this line — and often cross it:

• Mass-producing pages for every query variation? That's doorway abuse.
• Using AI to rewrite existing content in "AI-optimized" formats across hundreds of pages? If the rewrite doesn't add genuine value, that's scaled content abuse.
• Creating content at scale specifically to appear in AI features rather than to serve users? That's exactly what Google's policies target.

Google's spam documentation at developers.google.com/search/docs/essentials/spam-policies lays this out clearly. The consequences aren't theoretical — sites hit by spam actions can lose rankings entirely, including from AI features.

Quality Rater Guidelines on AI Content

Sections 4.6.5 and 4.6.6 of Google's Quality Rater Guidelines specifically address AI-generated content. The guidance isn't a blanket ban — Google is clear that AI-generated content is acceptable if it provides genuine value. But low-effort AI content that exists primarily for search visibility is flagged as low quality.

This matters because many GEO services rely heavily on AI content generation. "We'll create fifty AI-optimized pages for your brand" sounds productive until Google's quality raters (and increasingly, their automated systems) evaluate that content and find it's thin, repetitive, and adds nothing new to the topic.

Google's Position on AI-Generated Content

Google published a specific guide on using generative AI for content creation. The key points:

• Method of creation doesn't matter — Google evaluates the output, not the process
• Quality is the only criterion — AI content that's genuinely helpful ranks. AI content that's low-effort spam gets penalized.
• Mass production without value = spam — Using AI to scale content production without adding genuine expertise, original information, or unique value violates spam policies

The implication for GEO services that rely on AI-generated content at scale is clear: if the content isn't genuinely valuable on its own merits, the strategy is a ticking time bomb.

So What Should You Actually Do?

If you're an agency owner or marketing leader being pitched GEO services, here's a practical framework for evaluating what's worth your budget.

Ask this question about any proposed tactic: "Would this still be good practice if AI search features didn't exist?"

If the answer is yes — better content, stronger E-E-A-T signals, cleaner technical SEO, genuine reputation building — it's worth doing. It will help you in traditional search AND AI features, because they use the same systems.

If the answer is no — llms.txt files, AI-specific formatting, mention farming, content mass-production — you're paying for something that either doesn't work or actively risks a spam penalty.

The real competitive advantage in AI search isn't a secret tactic. It's the same advantage it's always been: produce content so good, so original, and so trustworthy that Google's systems — whether serving traditional results or powering AI features — consistently choose to cite you.

That's not a sexy pitch. It doesn't fit neatly into a new service package with a three-letter acronym. But it's what Google's own documentation says, and it's the only strategy that will hold up as AI search continues to evolve.

The Real Problem GEO Should Be Solving

Here's the thing — the underlying concern driving GEO adoption is completely valid. AI platforms are reshaping how buyers discover and evaluate brands. When someone asks ChatGPT "what's the best project management tool for agencies?" and your product isn't mentioned, that's a real business problem. When Perplexity tells a potential customer that your competitor is the market leader and doesn't mention you at all, that matters.

The problem isn't awareness of this shift. The problem is the response.

The GEO industry's response has been to invent new manipulation tactics — llms.txt files, content chunking, mention farming. Google's own documentation tells us these don't work, and some of them risk penalties.

The right response is a combination of two things:

First, visibility and diagnosis. You need to know what AI platforms actually say about your brand today. Not what you hope they say. Not what your marketing team assumes. The actual responses, across multiple platforms, for the queries your buyers use. Most brands are flying blind here — they have no idea whether AI recommends them, ignores them, or actively sends buyers to competitors. Beyond just monitoring, you need to understand why. When a competitor gets recommended and you don't, what sources is the AI drawing from? What content do they have that you're missing? Are their citations built on solid authority, or are they riding unstable tactics that could collapse with the next algorithm update? That diagnostic layer — tracing AI citations back to their sources and identifying the specific evidence gaps — is where the real insight lives.

Second, targeted action using fundamentals. Once you can see exactly where you're invisible and exactly why the AI is choosing competitors over you, the fix isn't a GEO trick. It's targeted, specific work informed by that diagnosis: creating the original content that fills your evidence gaps, building the E-E-A-T signals you're missing, strengthening the specific citation sources where you're weak. The strategies are fundamentals — the same work Google has always rewarded — but the targeting is what makes it efficient. You're not doing generic SEO and hoping for the best. You're closing specific gaps that your diagnosis uncovered.

The brands that will win in AI search aren't the ones buying GEO packages. They're the ones that can see the problem clearly — and then do the hard, legitimate work of becoming the source AI platforms want to cite.

That starts with knowing where you stand. You can't fix what you can't see.