How AI search actually works (and why it's not SEO)
You can't optimize for a system you don't understand. So before a single tactic, let's open the hood. You don't need to be technical — you need an accurate mental model, because almost everything people get wrong about AI search comes from quietly assuming it works like Google. It doesn't.
The two ways an AI answers you
Mode one: from memory (training data). Models like GPT, Claude, and Gemini learned from an enormous slice of the public internet up to a cutoff date. If your business and your content were repeated enough across enough sources, the model "knows" you — it can name you with no internet connection. This is the deep game, and it's slow.
Mode two: from live retrieval (RAG). The AI takes your question, runs live searches, pulls back a handful of current web pages, reads them, and writes an answer grounded in what it just fetched — usually with citations. Perplexity works this way by default; Google AI Overviews do it on nearly every query. This is the game you can win in weeks, not years, because it rewards being the clearest, most retrievable source on the live web right now.
Why this isn't just SEO with a new coat of paint
- There's often no click. Your win condition is being named, quoted, or cited — even when the user never visits your site.
- The unit is the passage, not the page. AI systems pull specific paragraphs, lists, definitions, stats. You're writing extractable, self-contained chunks.
- Being cited beats being ranked. A confident, sourced, well-organized page from a small brand can get cited over a vague page from a huge one.
- The competition includes synthesis. Make your content so clearly the best source that pulling it is easier than making something up.
The retrieval pipeline, end to end
- The query gets interpreted and often rewritten. "Who can make my coffee brand look good" might become "branding studio for coffee companies."
- A search runs against an index — sometimes the engine's own, often Bing or Google.
- Pages get fetched and read. Slow, blocked, or JavaScript-buried pages are already out.
- Passages get selected — the chunks most relevant and trustworthy. This is where extractable writing wins or loses.
- The answer gets synthesized with citations. Get selected here and you're named in front of the buyer.
Be retrievable. Be extractable. Be trustworthy. Be known well enough that you survive the query rewrite. That's the whole game.
Open ChatGPT, Perplexity, and Google, and ask each the five questions your best customer would ask when they're ready to buy but don't yet know you exist. Not "[your brand] reviews" — your customer doesn't know your name yet. Ask things like "best [your category] for [your niche]."
Write down, for each engine: Were you mentioned? Who was? What got cited? You now have a scoreboard. The rest of this course is about moving those numbers.
Next transmission — the citability mindset — arrives in your inbox tomorrow.