Module 1 — Foundations: GEO vs SEO vs "AI Search"

🧠This module gives you the mental model that everything else is built on.

Why this chapter matters most

There is a natural temptation when starting in GEO: jump straight to tactics. "Just tell me what to put in llms.txt and I am done." That is understandable, but it is also the fastest way to waste months optimizing what does not move the needle. The discipline is so new that almost nobody has a solid mental model of how it works under the hood yet — and that is precisely your advantage: if you understand the mechanism before your competition does, every action you take will have intention behind it instead of being a blind bet.

Think of the difference between someone who knows how to drive and someone who also understands how the engine works. Both reach their destination on a sunny day with a clear road. But the day something goes wrong — the car loses power on a hill, a warning light comes on — only one of them knows what is happening and what to adjust. In GEO, the "sunny days" are the months when everything seems to work without you knowing why; the "hill days" are when AI stops citing you, starts recommending a competitor, or describes your product incorrectly, and you need to diagnose the real cause. This module turns you into the person who understands the engine.

The goal of the next few pages, stated plainly, is for you to internalize one big idea and turn it into instinct: when an AI system looks at sources to answer, your content must be easy to find, easy to understand, and safe to cite. It sounds simple, and it is. But each of those three conditions hides a chain of technical and editorial decisions that will occupy the rest of the Academy. Today we build the frame where they all fit.

From ten blue links to the single answer

For twenty-five years, "searching" meant the same thing to almost everyone: you typed something into Google, it returned a list of links, and you chose which one to click. All of classic SEO — keywords, backlinks, meta titles, page speed — was built around that ritual. The game was to be one of those ten blue links and, if possible, the first one, because the top result captured most of the clicks.

That ritual is breaking apart before our eyes. More and more, people do not want ten links: they want the answer. They ask an assistant and receive a written paragraph, a ready-made comparison, a recommendation with a name attached. In many cases there is not even a follow-up click: the conversation starts and ends inside the assistant. When that happens, the strategic question stops being "Do I appear in the list?" and becomes "Am I the source the assistant decided to summarize and cite?"

That shift — from the link list to the generated answer — is the full territory of GEO. It is not a minor SEO tweak; it is a game with its own rules. In SEO the end goal is the click; in GEO the end goal is to be the cited source inside the answer. And the signals that win each game only overlap in part, as we will see. That is why a site can rank first on Google and, at the same time, be practically invisible to ChatGPT.

SEO and GEO: relatives, not twins

It is worth saying this clearly to avoid two opposite mistakes. The first mistake is thinking GEO is "SEO by another name" and that you can keep doing what you have always done. The second is thinking SEO is dead and you should throw it all away. Neither is true.

SEO is still alive and still optimizes for a search engine's ranking algorithm: it wants position on the results page to earn a click. Its classic signals are keywords, inbound links, meta tags, and experience metrics like Core Web Vitals. GEO, by contrast, optimizes for how a language model reads, understands, trusts, and cites. Its signals belong to a different family: the extractable clarity of each paragraph, the existence of a recognizable entity behind your brand, structured data that says "what is what," an llms.txt that guides agents, and the reputation other sources build about you.

The most honest way to summarize it is this: SEO gets you to the results page; GEO gets you inside the answer. They do not compete; they stack. The work of this Academy is not for you to abandon your SEO, but to add the layer the era of generative answers demands. Keep this phrase handy, because you will use it often when someone asks, "Isn't this just the same old SEO?"

What "being cited" really means

When someone starts in GEO, they usually reduce success to "my URL shows up." That is a weak definition. Being cited, in the sense that matters to us, is something deeper: the system treats your page as a source of truth — it trusts that your content answers the question correctly and is willing to lean on it to build its answer, mentioning you as the origin.

There is a gradation worth keeping in mind. At the lowest level, the assistant does not even consider you: you do not enter the pool of candidate sources. One step up, you enter the pool but the system chooses to read and cite someone else. Higher still, it reads you and uses you, but does not attribute you. At the top, it reads you, leans on you, and cites you by name. Each of those jumps corresponds to a different kind of work you will do across the Academy: entering the pool is mostly discovery and architecture; being chosen is mostly clarity and structure; being cited with confidence is mostly entity and authority.

There are also three moments of truth that will repeat like a refrain throughout the program, and they are worth memorizing now. First: Do you enter the candidate pool? Second: Are you chosen to be read? And third: Are you cited? Every module in this Academy, without exception, attacks one of those three moments. When you wonder why you are doing a task, ask which of the three moments you are trying to win; if you cannot answer, that task is probably not a priority.

The three ways AI uses your content

To optimize with judgment, you need to distinguish the three paths by which your content can end up influencing an AI answer, because you do not control them all equally.

The first is pre-trained knowledge. The model "remembers" things it learned during training, from a huge corpus captured in the past. If your brand was known and well described when the model was trained, some of that will remain in its memory. But this path has two serious problems: you do not control it directly, and it is always outdated, because training has a cutoff date. Never trust that the model "remembers" your current prices: it will not, and if it tries, it will probably get them wrong.

The second path is retrieval, technically known as RAG (retrieval-augmented generation). Here the system, instead of relying only on memory, goes out to find fresh information to ground its answer. It is the most important path for GEO because it is where your current content enters play in near real time. And it has one decisive technical detail worth understanding well: when an assistant "searches," it does not retrieve whole pages — it retrieves fragments (called chunks) selected by semantic similarity to the question. That has an enormous practical consequence: self-contained paragraphs win — those that answer on their own without needing the rest of the page. A paragraph that starts with "as we said before..." is a bad chunk, because out of context it says nothing. Keep this idea: much of the citable content work in Module 4 is, at bottom, writing good chunks.

The third path is live browsing through agents: systems that visit URLs in the moment, read what they find, and act accordingly. Here it matters enormously what the agent receives when it arrives (your HTML full of menus and scripts, or a clean, readable version?) — and whether there are signals that guide it toward what actually matters. Modules 6 and 7 live in this territory.

Your GEO work concentrates, above all, on the second and third paths: retrieval and browsing. The first, pre-trained knowledge, you influence only indirectly and over the long term, by building an entity so solid and so repeated by reliable sources that future training runs absorb it. But that is not where you will see results in weeks; it is a consequence, not a direct lever.

The full picture: how an assistant thinks when asked about you

It is worth walking through, step by step and in slow motion, what happens when someone asks an assistant a question about your category. Imagine a user types: "How much does BotPass cost and what does it include?"

First, the system understands intent. It does not stop at the literal words: it interprets that the person wants pricing and scope for a specific product, probably to compare or decide on a purchase. That nuance matters later, because it determines what kind of content it will consider a good answer.

Second, it decides whether to answer from memory or search. A good system knows prices change and that trusting memory is risky. So it decides to search for an up-to-date source of truth instead of risking an old figure. This moment is key: the more "volatile" a data point is (prices, availability, terms), the more likely the system is to retrieve instead of relying on memory — and therefore the more weight your current content carries.

Third, it chooses which sources to read. Here everything you will work on in the discovery and authority modules comes into play: whether your page exists, whether it is findable, whether your brand is recognized as an entity, whether other sources back you up. The system assembles a small candidate set and prioritizes.

Fourth, it reads and extracts. The assistant enters the chosen pages and takes what it can understand. And here is where a huge number of sites lose without realizing it: if your page is full of menus, banners, scripts, pop-ups, and filler text, the model receives noise and struggles to find the signal. If instead it finds a clear definition, well-defined plans, explicit terms, and a link to the canonical source, it extracts exactly what it needs.

Fifth, it drafts and, sometimes, cites. The system composes an answer from what it extracted. If your page was clear and reliable, it is likely to lean on it and mention you. If it was confusing, it may mix your plans, misattribute a term, or — worst of all — fall back on an external source worse than yours simply because it was more readable.

The moral of the story is uncomfortable but liberating: you cannot control the question, but you control almost entirely what the assistant finds when it lands on your site. And that landing is where the citation is won or lost. When an assistant reaches your page, it should be able to find within seconds what your product is, what its plans or terms are (if applicable), what its limits and exceptions are, and a clear link to the canonical source where the official truth lives. If that is there, you play with an advantage. If not, you depend on luck.

Not all bots come for the same reason

One nuance that saves a lot of misunderstanding: when we talk about "AI bots," we are not talking about one single thing. It helps to distinguish at least three families by intent, because they do not matter equally to you.

There are discovery and navigation bots, which come to read in order to answer a specific query right then or shortly after. They are your best friends: they are the direct path to being cited, and they are the first ones you should optimize for.

There are collection and training bots, which capture content at scale to feed future models. Their effect is slower and more diffuse: they contribute to that pre-trained memory we discussed, but they do not give you a citation tomorrow. Optimizing for them matters long term, not urgently.

And there are scrapers and third parties, which are often simply noise, and sometimes abuse. Here the decision is less about optimizing and more about deciding what you allow, what you limit, and what you monetize — a topic you will address in Modules 6 and 10.

The good news is that you do not have to guess which of these families visits you or how often. That is exactly the information BotPass puts in front of you, and it is what we dedicate all of Module 2 to.

Where BotPass fits in all of this

Until now we have talked about mechanisms. Now, the tool. BotPass is what turns this conceptual frame into something measurable and actionable, which is why you will use it in every module.

At the most basic level, BotPass shows you which bots are reading you and which URLs they consume, so you stop guessing what to optimize. Without that data, all GEO work is a bet; with it, it is a plan. It also helps you identify your sources of truth — those pages agents are already using to answer, such as pricing, documentation, or policies — and supports a mindset of transparent access: instead of blocking everything out of fear, you decide where you want to be perfectly readable, where you want to guide the agent toward what matters, and where, if needed, you monetize consumption — all without changing the underlying truth your human visitors see.

On top of that foundation, BotPass adds two metrics that anchor the Academy and that you will start tracking today. The first is the GEO Score, a 0–100 audit spread across six dimensions — frontmatter, structure, schema, discovery, content, and citability — with a letter grade from A to F. It is your starting diagnosis. The second is the Entity Equity Score, the flagship metric that combines into a single number your entity recognition, your schema, your verified profiles with sameAs, the consistency of your identity (NAP), the coherence of your content clusters, and third-party mentions. It measures your real authority with AI over the medium term. You will watch both rise throughout the program — and that rise is, literally, your progress.

Mini-method for today: your first three moves

Theory only sticks if you land it on your own site today. Before moving to Academy Level, spend an hour on these three moves. They are not optional: they are the raw material you will work with in the nine modules that follow.

The first move is to define your three to five main topics. A main topic is a "question territory" you want to win inside AI — not a lone keyword, but a space of intent. Think of things like "AI scraping," "content licensing," "llms.txt," or "GEO for ecommerce." Write three to five. If you write fifteen, you lack focus; if you write one, you lack ambition. The three-to-five range is where real work happens.

The second move is to list your entities. For each topic, note your product or service, its features, its plans, its use cases, and its alternatives or competitors. This list looks bureaucratic, but it is gold: it is the vocabulary AI must learn to recognize you by, and it will be the foundation of the entity work in Module 5.

The third move is to choose your twenty source pages. And watch the criterion, because it is counterintuitive: you do not choose the prettiest pages or the ones you like most — you choose the ones that should be cited when someone asks about your category. Fill in a table like this, which will serve as your starting inventory:

When you have these three things — topics, entities, and source pages — you will have something most people "doing GEO" do not: a map of what you are trying to win and which assets you have to get there.

What a solid GEO foundation looks like

So it does not stay abstract, look at how the top of a well-built "source of truth" page should read. The key is the answer first principle: the essentials up top, in a block a human understands in fifteen seconds and from which an agent can extract definitions, purpose, terms, and canonical links without inventing anything.

⚡TL;DR (Answer first)

This block works for a very concrete reason that connects to everything above: it is a perfect chunk. If a retrieval system takes only that fragment, it is still useful and correct out of context. It answers the "what," the "who for," the "how," points to where the official truth lives, and even dates itself so the model knows it is fresh. When in Module 4 we talk about writing to be cited, you will come back to this example mentally.