The AI search landscape

Search is becoming answer. When AI Overviews, ChatGPT, Perplexity, and Gemini synthesize a single answer, the job shifts from ranking to being the source the machine cites. This module maps the landscape, shows how LLM-powered search and RAG actually work, quantifies what AI answers are doing to traffic, and lays out the SEO/AEO/GEO stack and metrics the rest of the course is built on.

Why AI search changes everything

AI search replaces a list of links with a single synthesized answer, so the game changes from ranking to being cited. When Google’s AI Overviews, ChatGPT, Perplexity, and Gemini answer the question directly, the user often never clicks — which means your job is no longer to win position one, but to be the source the machine quotes inside its answer. That is the whole of AEO and GEO, and it is already mainstream, not a forecast.

For twenty years search marketing had one shape: produce a page, earn a ranking, capture the click. AI search breaks the third step. The engine reads the open web, synthesizes an answer, and hands it to the user with a handful of citations — if you’re lucky enough to be one of them. The click you used to compete for frequently never happens, because the user’s question was answered in place.

This is not a fringe scenario you can wait out. AI answers now sit on top of a large and rising share of searches, and a generation of users is learning to ask ChatGPT or Perplexity instead of opening ten tabs. The brands that adapt early — structuring content to be quotable, building the authority machines trust, and measuring citation instead of just rank — are the ones that will still be visible when the link economy finishes shrinking.

Claim: Google’s AI Overviews reached about 2 billion monthly users by mid-2025, and now appear on roughly half of tracked search queries. Source: TechCrunch / industry trackers. Context: AI answers are no longer an experiment bolted onto Search — for a large share of queries they are the default surface a user sees first.

From ten blue links to one answer

Search has moved through four eras: keyword-matching (1998), the knowledge graph and featured snippets that began answering questions on the page (2012–2016), the AI Overviews / Search Generative Experience era (2023–2024), and the standalone answer engines — ChatGPT Search, Perplexity, Gemini — that bypass the traditional results page entirely (2024→). Each step moved the answer closer to the user and the click further away.

Knowing this arc matters because every ‘new’ AI-search tactic is really the latest move in a twenty-year drift from links toward answers. Featured snippets taught us to write a 40-word direct answer; AI search simply raised the stakes of that same skill. Click the timeline to see how steadily the click has been disappearing — and why ‘zero-click’ was a trend long before ChatGPT.

Claim: In SparkToro’s 2024 zero-click study, for every 1,000 US Google searches only about 360 clicks reached the open web — a majority of searches ended without a click even before AI answers fully scaled. Source: SparkToro — 2024 Zero-Click Search Study. Context: Zero-click is a long-running structural drift, not an AI novelty — AI Overviews accelerate a trend that began with answer boxes a decade ago.

The AI search landscape today

Five surfaces matter. Google AI Overviews — AI answers inside the results most people still use. Google AI Mode — a full conversational search experience. ChatGPT (with Search) — the largest standalone assistant, now answering live web questions. Perplexity — an answer engine built around citations. Gemini — Google’s assistant woven across its products. Each retrieves and cites differently, so ‘being visible’ means earning trust on several at once.

The trap is treating these as one thing. They draw on different indexes, weight sources differently, and cite differently — Perplexity is citation-first and surfaces its sources prominently, while ChatGPT may synthesize from training plus live retrieval and cite more sparingly. Winning on one does not guarantee the others. Click through to see what each rewards.

How LLM-powered search actually works

An AI search answer is built in three steps: retrieve (find candidate passages relevant to the query, often from a live web index), rank/select (decide which passages to trust and use), and generate (write a fluent answer grounded in those passages, with citations). You can influence all three — but the highest-leverage point is retrieval and selection, because a passage that is never retrieved can never be cited.

This is the engine-room knowledge that separates real practitioners from people repeating ‘just write good content.’ The model isn’t reaching into a magic memory; for current questions it is running a retrieval system, pulling passages, and grounding its answer in them. If your content isn’t crawlable, isn’t chunked into clean retrievable passages, or isn’t trusted enough to be selected, you are invisible no matter how good the prose reads to a human.

1. RetrieveThe engine converts the query into an embedding and finds semantically similar passages from its index (and/or live web fetches). If your page can’t be crawled or rendered, it isn’t in the candidate set.
2. Rank & selectAmong candidates, the system favors passages that are clear, self-contained, sourced, and from trusted entities. This is where E-E-A-T and structure decide whether you make the cut.
3. Generate & citeThe model writes a grounded answer and attaches citations to the passages it leaned on. Quotable, unambiguous passages are far likelier to be the ones quoted and linked.

RAG, embeddings, and why retrieval is the battleground

Retrieval-augmented generation (RAG) is how AI search stays current: instead of relying only on what the model memorized in training, it retrieves fresh passages at query time and grounds its answer in them. Embeddings are the math that makes retrieval work — text is converted into vectors so the engine can find passages by meaning, not exact keywords. The practical takeaway: write the way the question is meant, in clean self-contained chunks, because that is what gets matched and pulled.

Two consequences fall out of this and they reshape content strategy. First, relevance is now semantic — you no longer need the exact keyword, you need to clearly mean the same thing, so natural, specific language beats keyword-stuffing decisively. Second, retrieval happens at the passage level, not the page level: the engine pulls a chunk, not your whole article. A brilliant page where the answer is buried in paragraph nine will lose to a mediocre page that states the answer cleanly in a self-contained block near a relevant heading.

Claim: Princeton’s “GEO: Generative Engine Optimization” study (Aggarwal et al., KDD 2024) found content tactics could lift visibility in AI answers by up to ~40%, with adding statistics improving visibility ~41% and citing authoritative sources improving it ~115% for lower-ranked content. Source: Princeton — GEO: Generative Engine Optimization (arXiv 2311.09735). Context: These are the first peer-reviewed, measured GEO levers — structure, statistics, and citations — not folklore; Modules 3 and 4 operationalize them.

What AI search is doing to your traffic

Expect fewer, better clicks. AI answers absorb the informational queries that used to drive top-of-funnel traffic, so raw organic sessions fall — but the visits you still get are often later-stage and higher-intent, because the easy questions were answered before the click. The strategic error is to panic about the volume drop; the right move is to re-target measurement and content toward citation share and high-intent capture.

Be clear-eyed and avoid both denial and doom. Yes, informational traffic is eroding and zero-click is rising. But also: traditional search is still vastly larger than AI referral traffic today, and Google still dominates — so abandoning SEO for a pure ‘AI-only’ play is as wrong as ignoring AI entirely. The durable position is to defend classic search while building AI-search visibility on top of it, because the same assets (authority, structure, schema) feed both.

Claim: As of 2025, Google still held roughly 90% of global search and AI assistants drove well under 1% of referral traffic — even as AI answers reshaped what happens on the results page. Source: BrightEdge / industry analysis (2025). Context: The honest framing: AI search is changing visibility and clicks now, but classic search is still where most traffic lives — optimize for both, don’t abandon one for the other.

The strategic response: SEO, AEO, GEO

Three disciplines stack. SEO earns rankings and crawlability — still the foundation, and still how you get into most AI indexes. AEO (answer engine optimization) structures content to win direct answers and citations — clean Q&A, schema, snippet-ready passages. GEO (generative engine optimization) shapes how generative engines synthesize and cite you — statistics, sources, authority, and entity clarity. They are layers, not alternatives: GEO without SEO is a house with no foundation.

The rest of this course is built on that stack. AEO fundamentals (Module 2) and GEO (Module 3) are the two new layers; content structure (4), schema (5), and E-E-A-T (6) are the levers that power them; and measurement (7) replaces the rank-tracking you can no longer rely on. Treat them as one integrated practice and you stay visible across both classic and AI search; treat AEO/GEO as a bolt-on gimmick and you’ll chase tactics that don’t compound.

Is SEO dead now that AI answers everything?

No. SEO is how you get crawled, indexed, and trusted — the prerequisite for being retrievable by AI search at all. AI Overviews heavily favor pages that already rank. SEO changes shape; it doesn’t die.

What’s the difference between AEO and GEO?

AEO optimizes to win a direct answer or citation (structure, Q&A, schema, snippet eligibility). GEO optimizes how generative engines synthesize and cite you (statistics, authoritative sources, entity authority). AEO is largely a superset of snippet/answer tactics; GEO adds the generative, multi-engine layer.

Should I stop tracking keyword rankings?

Don’t stop, but stop treating them as the whole story. Add citation/mention tracking across AI engines and high-intent capture, because a page can lose its click while still ‘winning’ the answer (Module 7).

The metrics that replace rankings

When the click disappears, rank-tracking alone goes blind. The new scorecard measures citation share (how often AI engines cite you for your priority questions), presence (are you mentioned in the answer at all, cited or not), share of voice vs competitors across engines, and downstream high-intent traffic and conversions from the clicks you still earn. You measure these by querying the engines and by watching analytics for AI-referral and branded-search lift — the subject of Module 7.

The mindset is the hard part. A page can lose half its clicks and still be a strategic win if it became the cited authority an AI quotes to millions — brand presence inside the answer compounds even when the click doesn’t happen. Conversely, a page that still ‘ranks’ but never gets cited is quietly losing the new game. You manage what you measure, so the scorecard has to change before the strategy can.

Five ways brands misread AI search

The predictable errors: panicking and abandoning SEO, treating all AI engines as one, optimizing prose humans love but retrievers can’t see (JS-only content), chasing AI-referral traffic numbers that are still tiny instead of citation share, and bolting on ‘GEO tricks’ without the authority and structure that make them work. Each comes from misunderstanding that AI search is a retrieval-and-trust system, not a magic box.

Your AI-search readiness checklist

AI-search readiness is a finite list. Work down it; the early items make the later ones possible. Tick only what is genuinely true of your site today.