What this chapter does to you: "AI is replacing SEO" is a slogan, not an accurate model of what's happening. What's actually happening is that a second and third discovery surface grew up alongside search rankings, each with its own retrieval mechanism and its own content shape. Optimizing for one doesn't automatically win the others. This chapter draws the actual boundaries so you can tell which game you're playing at any given moment.
What each one actually optimizes for
| SEO | AEO | GEO | |
|---|---|---|---|
| Wins | A ranking position among ~10 blue links | Being the single direct answer (featured snippet, People Also Ask, voice assistant) | Being one of the ~2–7 sources an AI system cites when synthesizing an answer |
| User behavior | Clicks through to your site | Often doesn't click — consumes the answer in place | Never sees your URL at all unless the AI cites it inline |
| Retrieval mechanism | Search index, ranked per query | Search index, extracted as a direct-answer candidate | Either live web retrieval (Perplexity, ChatGPT browsing, Google AI Overviews) or a model's training data or a fixed RAG corpus |
| Content shape that wins | Comprehensive, well-linked, technically healthy | Concise, question-answering, structured with clear headers | Fact-dense, quotable, front-loaded with the direct answer |
The retrieval mechanism row is the one worth sitting with, because it's where the real strategic difference lives.
Three different retrieval paths
GEO via live retrieval is the one you can actually influence with content work, and it behaves a lot like AEO — the AI is doing search retrieval under the hood, then synthesizing across sources instead of picking one. This is why a page that's invisible to crawling (chapter 1) is invisible here too: live-retrieval GEO inherits every SEO fundamental, it just adds a synthesis step on top.
GEO via training data is a different, much slower game — being prominent, consistent, and repeated enough across the web that you show up in a model's general knowledge without any live citation. You can't "SEO" your way into a training run; the closest lever is sustained, widely-echoed presence over a long time horizon.
GEO via a fixed RAG corpus is a special case: someone else built a retrieval system over a bounded set of documents (their own product docs, a customer support knowledge base) and you have no influence over it unless your content is literally in that corpus. Not usually the target of a public-facing SEO/GEO strategy, but worth knowing it exists as a category — you'll recognize it immediately from the RAG chapter in AI Engineering, because it's the exact same architecture, just now something you're being retrieved by instead of building.
Why the caveats on AEO changed recently
FAQPage structured data used to be a reliable way to win rich results in
Google Search. That eligibility has since been restricted to a narrower
set of authoritative/government-style sites — don't build a content
strategy around FAQ schema expecting the same payoff articles from a few
years ago describe. This is a good example of why grounding this section in
current sources mattered: SEO/AEO tactical advice has a short shelf life,
and stale tactics don't just underperform, they can actively waste
engineering time implementing something that no longer does what it used to.
The layering, not replacement
The instinct when a new channel appears is to chase it first. The more useful framing, borrowed from how experienced practitioners actually sequence this work: fundamentals, then trust, then AI-specific optimization — in that order, not the reverse.
The reasoning is structural, not just conservative: every layer above depends on the one below it working. AI-specific tactics applied to a page that isn't even crawlable (chapter 1) do nothing — there's no page for the tactic to act on. Fact-dense, quotable content on a domain nobody trusts or references elsewhere has little to be "cited" against. Skipping straight to layer 3 because it's the newest, most-talked-about layer is the single most common mistake technical teams make here — it's optimizing the top of a stack whose bottom hasn't been built yet.
The next two chapters go one level deeper into layers 1 and 3 respectively: chapter 3 is the concrete technical-SEO implementation (layer 1, made real in your actual Next.js stack), and chapter 4 is content structuring specifically for GEO (layer 3). Chapter 5 comes back to tie the whole stack into an actual startup content strategy.