Three steps between your page and an AI answer
Before any AI can mention your brand, three things must happen in sequence: a crawler must fetch your page, a retrieval system must surface it as relevant to a query, and the language model must decide to cite it in its response. A failure at any stage means your content disappears from the answer — silently, with no error message and no ranking drop to diagnose.
Step 1: crawling — can the bot even read your page?
AI crawlers are not browsers. They send an HTTP request, download the raw HTML response, and move on. They do not run JavaScript, load fonts, trigger lazy-image loaders, or wait for client-side rendering to complete. If the content a visitor sees on your page is assembled in the browser — by a React framework, a headless CMS hydrating after load, or a JavaScript product configurator — an AI crawler sees none of it. This is called the render gap, and it is one of the most common reasons a well-optimised site still earns zero AI citations.
Step 2: retrieval — does the model know your page exists?
Large language models do not query the live web for every question. Most answers come from training data (knowledge baked in during model training) or from a retrieval-augmented generation (RAG) layer that searches an index at query time. For your content to appear in training data, it had to be crawled, processed, and included before the training cutoff. For RAG retrieval, it must be in the index the model queries — which is controlled by the platform (Perplexity, Bing, Google AI Overviews) not by you. Optimising for retrieval means being the kind of source those indices want: authoritative, well-structured, and freshly updated.
Step 3: citation — why the model names you (or does not)
Even when a model retrieves your content, it has to decide whether to cite your brand in the answer it composes. The signals that influence this decision are not the same as the signals that win SEO rankings. Keyword density is irrelevant — in controlled studies it slightly reduces citation rate. What moves the needle is whether your content is authoritative enough for the model to trust, clear enough for it to excerpt, and specific enough to be the best available answer to the exact question being asked. The rest of this course covers each of those levers in detail.