Module 3 — Information architecture for AI

🗺In this module you build the "map" that humans and agents use to reach your

Your site is not a list of pages, it is a system of routes

There is a mindset shift that unlocks everything in this module, and it is worth pausing on it. We usually think of a website as a collection of pages: the home page, the blog, pricing, contact — a list of rooms in a building. That is a comfortable way to think for a human navigating with a menu in front of them. But an AI agent does not experience your site as a list of rooms; it experiences it as a system of routes, paths that lead from a question to an answer. And what decides whether you get cited is not how many rooms you have, but whether there is a short, clear path from the user's question to the truth you hold.

The ideal route, the one you want to exist for every important topic, looks like this: question → answer-first page → source of truth → (optional) comparison or policy. Someone asks something; there is a page that answers immediately and clearly; that page points to the canonical source of truth where the official data lives; and, if needed, there is a link to a comparison or the policy that qualifies it. When that route exists, the agent follows it effortlessly and cites you with confidence. When it does not, something dangerous happens: the agent improvises. And improvisation, which is tolerable on an opinion topic, is exactly what you do not want on prices, policies, or critical claims — because improvising on your prices means AI invents them or pulls them from a worse source.

Why architecture wins or loses citations

Stop to think about the effort asymmetry. For an agent, a clear route is cheap to follow and a broken route is expensive to reconstruct. If your truth about "how much it costs" is spread across an old landing, a two-year-old blog post, and the current pricing page — all three saying similar but not identical things — the agent faces three competing signals. In the best case it picks one at random; in the worst, it mixes all three and produces an answer that is not true on any of your pages. Dispersion does not just reduce clarity: it actively introduces error.

This connects to what you learned in Module 1 about how AI retrieves fragments. If the same idea lives in five different, slightly contradictory chunks scattered across your site, none of those chunks is as strong as a single canonical chunk would be — and they cannibalize each other semantically. Architecture, at bottom, is the art of concentrating signal instead of diluting it.

Where BotPass fits: do not draw a map nobody travels

Before you redesign your site on a whiteboard, an important warning so you do not fall into the architect-in-love-with-the-blueprint trap. It is tempting to design a perfect, elegant, symmetric map… that turns out bots do not care about. GEO architecture must be informed by real consumption, and that is where BotPass anchors you to reality.

Use BotPass first to identify which pages bots already treat as entry points. You will almost always be surprised: it may be the home page, yes, but pricing, documentation, or certain key guides you never imagined were so visited usually show up too. Those real entry points are the foundations of your map. From there, design hubs and internal linking so those entry points lead to your canonical sources of truth in one or two clicks, not five. And pay special attention to one signal: if BotPass shows bots consuming a lot of a page that is not canonical — that old landing, that forgotten duplicate — do not celebrate it as traffic, treat it as an alarm. It is a leak: AI is feeding on a version you do not control. The correct response is redirect, merge, or at minimum put crystal-clear links to the canonical page.

Hubs: the chapters of your book

Here is the first piece of architectural vocabulary. A hub is the pillar page for a main topic. And it is important to understand that a hub is not simply "a long post." A long post is content; a hub is an entry point with a structural function. A good hub does three things at once: it defines the topic clearly so anyone — human or machine — understands what it is about; it links in an orderly way to subpages that develop each facet of the topic; and it makes unmistakably clear which URL is canonical for each subtopic, so there is no doubt where the truth lives.

The metaphor that works best is the book. Hubs are the chapters of your site, and clusters — subpages grouped under each hub — are the sections within each chapter. A well-organized book has a few clear chapters, each with logical sections, and an index that takes you to any idea in seconds. A poorly organized book has the same information scattered without order, and even if the content is excellent, finding something specific is a nightmare. Your site, for AI, is exactly that: a book read by jumping around, and hubs are your index.

Sources of truth: one truth, one URL

Within each chapter there are pages with special status: sources of truth, or single source of truth. A source of truth is the page you want AI to cite when someone asks about a price, a policy, a definition, terms, or a methodology. These are the data points where accuracy is non-negotiable.

The rule here is brutally simple: if it is critical, it cannot be spread across five URLs with different wording. Every critical truth in your business should have a single canonical home, and everything else that talks about that topic should point to that home instead of repeating the data in its own words. This is not just content hygiene; it is a risk decision. Every duplicate of a price is a future contradiction waiting to happen the moment you change the price in one place and forget the other. And AI, consuming both, will propagate that contradiction multiplied. Consolidating your sources of truth is, at once, GEO work and risk management.

Internal linking for agents, not just SEO

Here is a subtle turn that distinguishes a GEO practitioner from a classic SEO professional. Traditional SEO thinks about internal linking in terms of ranking: which links distribute "authority" and help you rank. That logic is legitimate, but it is a different game. GEO thinks about internal linking in terms of understanding and citation: which links help an agent understand the structure of your knowledge and reach the correct truth.

That gives rise to a type of link that in SEO would be almost redundant but in GEO is gold: intent links. Instead of a generic menu, think in explicit signals like "if your question is about pricing, go to the canonical pricing page," "if your question is about terms, go to the canonical policy," "if you want to compare, go to the alternatives page." They do not have to be written that literally on the page surface — though sometimes it helps — but your linking structure should make that intent-to-destination mapping obvious. You are, in essence, leaving breadcrumbs for the agent so it does not have to guess.

Build your GEO Map v1: the step-by-step method

With the concepts in place, let us build your first map. It is a focused session of work, and the result — GEO Map v1 — is one of the most valuable documents you will produce in the entire Academy.

Start by defining your hubs. Take the three to five main topics you already identified in Module 1 and create one hub per topic. No more: if you end up with eight hubs, some are probably subtopics of others, and your map will lose focus. Fewer hubs, clearer, always wins.

Next, design the clusters for each hub. Each hub should support between three and eight subpages covering the different facets of the topic: definitions, use cases, practical guides, comparisons, and related documentation or policies. The table below is the tool to capture this structure; pay special attention to the "action" and "target canonical URL" columns, because they turn a static inventory into a work plan:

The third step is to mark one to three sources of truth per hub. Not every subpage is a source of truth; only those that hold critical data. Identify them explicitly, because they will receive priority attention in the content and schema modules:

And the fourth step is to design at least ten reading routes. A reading route is a small recipe that goes from a real question to the sequence of pages that answer it: the question, the answer-first page that handles it, the source of truth that grounds it, and optionally the comparison or policy that completes it. Writing ten of these routes forces you to verify, one by one, that the important paths on your site actually exist and are short. When you discover a route cannot be completed because a page is missing, you have just found a piece of your backlog.

An illustrated map, start to finish

So the method does not stay abstract, let us follow it with a concrete example. Suppose one of your main topics is "AI scraping." Your hub would be the pillar page /ai-scraping, which defines the topic and serves as the entry point. From it would hang several clusters: /ai-scraping/what-is with the canonical definition, /ai-scraping/how-to-detect as a practical guide, /ai-scraping/pricing-impact as a data benchmark, and /ai-scraping/alternatives as a comparison. Your sources of truth for this topic might be /pricing, where plans and terms live, and /policy/bot-access, where the bot policy lives.

Now observe how a concrete reading route materializes on this map. Someone asks: "How do I detect AI bots on my website?" The answer-first page is /ai-scraping/how-to-detect, which answers immediately; the source of truth that grounds it is /docs/bot-detection; and the policy that qualifies it is /policy/bot-access. The agent enters through the answer, confirms in the source of truth, and finishes at the policy — all in a clean sequence. That structural clarity is exactly what reduces ambiguity and increases the probability of a correct citation: the system "sees" a sharp route and follows it instead of improvising.