I’m gonna set the record straight before you pay another agency to “Optimize your content for AI/LLMs.” And here it is: LLMs don’t favor one fucking thing.
LLM’s don’t work like Google, or other search engines where pages are indexed. They’re not databases with a ranking algorithm. They don’t decide what site’s content gets surfaced. An LLM is a next token predictor, it reads a prompt, and statistically guesses what comes next based on patterns baked in during training. That’s it, it’s a one trick pony.
So when an agency or expert pitches you on GEO, AI SEO, or LLM Optimization, then they’re likely selling you something that doesn’t exist.
What’s Actually Happening Under the Hood of LLMs?
During training, LLMs are fed an enormous amount of text from the web. Books, forums, articles, Reddit threads, your competitor’s blog that’s mid… all of it. But here’s what the model doesn’t do: it doesn’t bookmark anything. It doesn’t store URLs. It doesn’t remember where it read what. What’s left after training is a frozen statistical snapshot of language patterns. It’s not a living index, and it’s not a database you can try to game.
Think of it like this: the model absorbed the internet the way you absorb years of reading. You don’t remember every book, magazine, or newspaper, but you develop a sense of how things work. So when an LLM is asked a question it doesn’t start “Looking it up.” Instead, it just responds, based on what it’s already internalized. That’s a really boiled down way of explaining how an LLM works.
So Where Does Retrieval Come In?
Here’s where it gets more nuanced, and where a lot of the “AI optimization” and GEO conversation gets conflated.
Some AI systems layer a retrieval mechanism on top of the base model. That’s called RAG (Retrieval-Augmented Generation). The retrieval part is actual Information Retrieval (IR): crawling, indexing, pulling relevant documents, then feeding them into the model to ground its response.
That retrieval layer is the part that resembles search. And guess what powers most of it? Search engines. Google. Bing. Web crawlers doing the same job they’ve always done. So when AI tools do surface your content, it’s not because the LLM has a soft spot for you. It’s because:
- A search engine indexed your page, and
- A retrieval system pulled it in as relevant context
The model itself is just the engine that synthesizes a response. It doesn’t pick winners. The index does.
What Does This Mean for Your Marketing Budget?
If you’re a marketing director or a business owner who outsources content, pay attention here because this is where money gets wasted.
“Optimizing for AI” at the base model level is not a service. It’s a sales pitch wrapped in buzzwords. The model is frozen. You can’t GEO, AEO, or whatever other acronym your way into its weights after the fact. What’s done is done.
What you can do, and what actually moves the needle, is this:
- Optimize for search engines. Fucking shocker, I know. If retrieval-based AI is going to cite you, it needs to find you first. That means the fundamentals: crawlability, authority, relevance, structure. Classic SEO. Not glamorous, but it works.
- Earn third-party coverage. Long before LLMs hit the scene, you’ve probably heard me say something like “Your brand should be everywhere your audience has conversations about your category.” And that’s because the model was trained on the web. If authoritative sources like trade publications, industry blogs, and credible outlets are talking about you, the model likely encountered that during training. You become part of the pattern. That’s brand presence at the model level. It’s slow, it’s hard, and it’s exactly why it matters.
The Bottom Line
The LLM/AI marketing landscape is full of people repackaging old snake oil with new labels. “LLM-optimized content” is the new “guaranteed Page 1.” It sounds plausible enough that busy marketing directors nod along in meetings, and specific enough that agencies can charge for it.
But don’t buy into that shit.
The real play hasn’t changed: build content that earns links, build a brand that earns press, create content that gets shared, make sure search engines can actually find and understand your stuff, and build a fucking brand that’s worth talking about, worth it to remember, and worth searching for by name. The AI tools that matter are mostly just retrieval systems in a trench coat, and retrieval runs on search.
If you’re old enough to remember the movie “Field of Dreams” it’s like that… “Build it and they will come.” If you do the fundamentals well, the AI visibility follows.
For a lack of better terminology, I wrote a post on “How to Optimize for AI Visibility.” But really, it’s just all the multichannel marketing efforts you should be focusing on if you want to show up on LLMs. Some of these “GEO strategies” can be manipulated/gamed, and some are more important than the others.
