Structuring internal linking so AI finds your key pages

AI crawlers find your key pages by following internal links, exactly the way a classic search engine does. The short answer: make sure every important page is reachable through descriptive anchor text in the raw HTML, group related content around hub pages and keep your central pages shallow. Then a model not only discovers that a page exists, but also understands what it is about and how it fits into your larger whole. In this article you will read how to tune your internal linking structure to that end.

This topic is a building block of the complete GEO guide. Where content architecture for AI extraction is about the structure within a page, this piece is about the structure between your pages.

How do AI crawlers discover your key pages?

AI crawlers discover pages by following links from pages they already know, starting at your homepage and other heavily linked entry points. There is no separate list that magically hands a model all your URLs. A page that no internal link points to is an island: technically online, but in practice almost impossible to find for a crawler that moves from link to link.

That makes your internal linking structure the mapping system of your site. Every link is a signpost that says: here is another relevant page, and this is what it is about. The clearer that network, the more completely an AI system maps out your site. The principle is the same as with classic website architecture for SEO, but with AI there is an extra constraint that many sites underestimate.

Why do AI crawlers often not see JavaScript links?

Most AI crawlers do not execute JavaScript, so links that only appear after loading via a script do not exist for them. Research into the crawl behaviour of bots like GPTBot, ClaudeBot and PerplexityBot shows that they fetch the raw HTML but do not run the scripts on it. What they see is therefore the page as it comes from the server, not the page after your browser has assembled everything.

The result is a silent failure. A navigation menu, a “related articles” block or pagination built by JavaScript looks perfect to you. For a non-rendering AI crawler there is simply nothing: no anchor tags, no path to the rest of your site. A site that looks fully connected to Google can fall apart, for an AI bot, into separate, unreachable islands.

The solution is simple in principle. Deliver your most important internal links as ordinary <a href> elements in the server HTML, not as something that only surfaces after a click or a script. To be sure, view the source code of a page (the raw HTML, not the rendered version in the inspector) and check that the links to your key pages are literally in there.

What is good anchor text for an AI model?

Anchor text is the visible, clickable text of a link, and for an AI model it is the most important signal about where that link leads and why. “Read more” or “click here” tells a model nothing. An anchor like “internal linking for AI” or “our approach to AI search” names the entity and the topic of the destination, before the model has even visited the target page.

For your internal links, that concretely means:

• Describe the destination, not the action. Use text that summarises the topic of the target page, not the act of clicking.
• Be consistent in how you name entities. When you refer to a service or concept, always use the same core term. That consistency strengthens how a model recognises and links the entity across your whole site.
• Avoid one anchor for ten different pages. If “more info” leads to ten destinations, every signal fades.
• Write for people first. A natural, readable sentence with a descriptive anchor works for your visitor and for the model. Do not force a search term where it does not belong.

Descriptive anchors also help an AI system with grounding, the process by which the model determines which source belongs to which claim. A clear anchor makes explicit which part of your site contains the answer to a given question.

How do you use hub pages to group entities?

A hub page is a central page about a broad topic that links to underlying, specific pages and also receives links back from them. For an AI model, such a hub is a strong signal: it shows which pages belong together, which entities are related and which page is your central answer on that theme.

Think of a hub-and-spoke structure. The hub covers the theme in breadth and points on to sub-pages that each dig into one aspect. Those sub-pages link back to the hub and, where it makes sense, to each other. This creates a bounded cluster in which the coherence is immediately readable for a crawler. The model does not see ten separate articles, but a coherent field of knowledge with a clear centre.

That setup does two things at once. It concentrates internal link value on your most important page, and it makes explicit which concepts you consider to belong together. For a B2B site that wants to be found on a specific area of expertise, that is exactly the signal that counts. How this fits into a broader GEO approach for B2B you can read in that separate piece.

Why is link depth important for AI visibility?

Link depth is the number of clicks a crawler needs to reach a page from your homepage, and the deeper a page lies, the greater the chance that it is discovered late or not at all. A crawler has a finite budget of time and attention. Pages that sit high in the structure and receive many internal links are read sooner by a model and interpreted as more important.

Aim therefore for a shallow structure in which every key page is reachable within a few clicks of the homepage or an important hub. A page tucked away deep, behind several submenus or only via a series of clicks, sends two wrong signals: it is hard to reach and it looks unimportant.

In practice this means: from your homepage and hub pages, link deliberately to the pages that work hardest for you. Not every page deserves the same prominence. By keeping your commercially most important pages, such as your AI search optimisation, shallow and well linked, you steer both crawlers and visitors towards what matters for your business, rather than towards a vain high page count.

The short summary

A good internal linking structure for AI rests on four pillars. Make sure links are in the raw HTML, because most AI crawlers do not execute JavaScript. Use descriptive, entity-focused anchor text so a model knows where a link leads. Build hub pages that group related content and mark out your central page. And keep your key pages shallow, so they are definitely reached and read as important.

At Customer Impact we always look at your internal links in terms of leads and revenue, not vanity figures. We are a small, fast senior team that honestly names what is and is not realistic, without guarantees on positions that no one can promise. Want to know whether your site lets AI crawlers find the right pages? Book your free intake.