GEO is a scam, and Google says so itself

For a few months now, it’s been impossible to open LinkedIn without running into a post promising to “dominate ChatGPT’s answers” through GEO, Generative Engine Optimization. A new job title, new $2,000 courses, new consultants reinventing themselves overnight as generative AI experts.

The problem is that Google just published its own guide on the topic. And that guide dismantles, point by point, nearly everything sold as GEO. I read it in full. Here’s what it actually says, and why most of what’s sold under that name has no serious basis.

GEO: a word to sell an old recipe in new packaging

The premise of GEO is simple on paper: generative search engines (Google’s AI, ChatGPT, Perplexity) supposedly work differently from classic search, so you’d need specific techniques to get cited by them.

Except when you look at what Google actually recommends, you find exactly the SEO fundamentals we’ve known for fifteen years. Topical authority, clear content structure, real experience, original data. Nothing new. Just a marketing name stuck on existing practices, plus a few techniques that are actually counterproductive.

Here are the four myths Google’s guide takes apart one by one.

Myth #1: llms.txt files are indispensable

You may have been sold the idea that you absolutely need to create an llms.txt file, modeled on robots.txt, to “guide” generative AI toward your most important content.

Google is clear on this: it isn’t necessary. AI models read and interpret a site’s content exactly the way classic search does, through standard crawling and indexing. There’s no parallel channel reserved for generative AI that would need a dedicated configuration file. Time spent producing an llms.txt is time that returns nothing.

Myth #2: you need to fragment your content for AI

Another common claim in the GEO ecosystem is that there’s an ideal paragraph length, an ultra-fragmented structure of tiny blocks, so AI models can “digest” the information more easily.

Here too, Google flatly contradicts the idea. Its systems understand the context and nuance of a full text. There’s no magic format, no optimal sentence length, no universal chunking to follow. What matters is that the content is clear and well organized for a human reader. Everything else follows naturally.

Myth #3: buying brand mentions boosts AI visibility

This is probably the practice most sold as “advanced GEO strategy”: paying to get your brand mentioned across as many third-party sites as possible, on the idea that citation frequency would shape how generative AI perceives you.

Google is direct about this: the practice isn’t just ineffective, it can also conflict with its anti-spam rules. Google’s spam detection systems automatically filter out this kind of manipulation. In other words, not only are you paying for nothing, you’re taking a real risk with your domain’s reputation.

Myth #4: you should write in “AI-friendly” language

The last myth, and perhaps the most telling about the conceptual emptiness behind GEO: the idea that you should rewrite your content with a specific syntax, phrasing built for machines rather than for humans.

Google’s models understand natural syntax, synonyms and the semantic relationships between concepts. There’s no reason to sacrifice writing quality for a human reader in favor of artificial phrasing supposedly more readable by a machine. Content well written for a human is, by construction, already well understood by a language model.

The technical proof: Google’s NLP was never separate from generative AI

There’s an even stronger argument than Google’s guide itself: the history of its own search engine. The GEO premise rests on the idea that there are two distinct worlds, classic search on one side, generative engines on the other, with different rules of understanding. Except when you look at what Google has built over the past ten years, that separation never existed.

RankBrain, launched in 2015, is the first building block. It’s a machine learning algorithm that converts words into vectors to understand novel queries and detect relationships between concepts, rather than relying on plain keyword matching.

BERT, launched in 2019, then changes things far more deeply. It’s a model built on the Transformer architecture, able to understand the bidirectional context of a sentence, meaning the words that come before and after at the same time. And this is the central point: this Transformer architecture is exactly the same technology family that powers every major generative language model today, whether it’s the ones behind Google’s AI answers or consumer conversational assistants. BERT is a language processing model, not a generative one, but it’s built on the same fundamental block.

MUM, in 2021, then the more recent Gemini models, push that same logic even further with multimodal understanding, able to cross text, image, audio and video to handle complex queries. That’s the same technological lineage that now powers AI-generated summaries in search results.

What this means in practice is that there was never a technological break that would justify a separate discipline. Optimizing for BERT in 2019 was already optimizing for what would become the foundation of today’s generative engines. And it settles the matter: there’s no “special language for AI” to learn, classic search already understands natural language with the same mechanisms wrongly attributed exclusively to GEO.

What Google actually recommends: topical SEO, not GEO

Once the four myths are cleared away, what’s left is the guide’s real recommendation, and it’s nothing exotic: build topical authority. In practice, that means three things.

This is exactly what I’ve done from the start on my own projects, through what I call semantic clusters: a pillar article that covers the topic as a whole, surrounded by satellite content that digs into each sub-theme, all linked together through coherent internal linking. No llms.txt file, no robot language, just a content architecture built for a human looking for a complete answer.

Why the GEO myth caught on so much

There’s a simple explanation for the GEO craze: anxiety. Many sites have seen their classic search traffic drop since AI-generated answers started appearing directly in results. That anxiety creates a market for any solution promising a quick fix and a method name that sounds new.

The problem is that market filled up with people selling recipes with no real technical basis, leaning on perceived urgency rather than verified data. Google’s guide, by putting things in black and white, strips away much of that discourse’s legitimacy.

The good news is that the real answer is simpler and more durable than any GEO trick: solid, structured content, backed by real expertise. What worked before keeps working.