There’s a story going around that SEO is over and something entirely new has replaced it. That’s not what the data on the ground looks like. When you trace where a large language model actually got the sentence it just wrote about a company, you usually land on an indexed web page, a forum thread, or a news article — things that have been the raw material of search for twenty years. The retrieval layer changed. The supply chain didn’t.
That’s the practical premise behind the Cory Maki SEO approach to Generative Engine Optimization: Generative Engine Optimization (GEO) — the practice of getting your brand surfaced and cited inside AI-generated answers — is not a replacement discipline. It’s a layer that sits on top of competent search work. If the fundamentals underneath are broken, the GEO layer has nothing to stand on.
Why this matters now
For most of the last decade, the goal was a blue link in position one. Now a meaningful share of questions get answered before anyone clicks anything — in Google AI Overviews, in ChatGPT, in Perplexity. The unit of success shifts from ranking to citation: being the source the model pulls from and names.
That shift matters most for anyone whose buyers research before they talk to you. SaaS founders. Law firms. Public figures. Over a decade in reputation and search, the pattern I keep seeing is the same: the brands that get cited in AI answers are rarely the ones running clever prompt tricks. They’re the ones with clean, crawlable, deeply-covered content that was already doing well in traditional search — plus a credible presence in the places where opinions get formed.
What I’m seeing across AI search is that the models are conservative. They prefer sources that are easy to parse, corroborated elsewhere, and unambiguous about who wrote them. Those are SEO virtues. They just pay out differently now.
How the mechanism actually works
Strip away the mystique and an AI answer is usually built in three moves:
- Retrieval. The system runs one or more searches — often rewritten versions of your question — against an index. If you aren’t in that index, or your page is buried behind JavaScript, blocked, or slow, you’re out before the round starts.
- Selection. From the retrieved set, the model picks passages that directly answer the sub-question. Short, self-contained, clearly labelled passages win. Rambling introductions lose.
- Synthesis and attribution. The model writes an answer and, depending on the surface, names sources. Brands that appear consistently across multiple independent sources are more likely to be named than a single page making a lonely claim.
Here’s a concrete version. Say a founder asks an assistant which contract-review tools handle multi-party redlining. The system doesn’t reason from memory about your product. It searches, grabs a comparison article, a documentation page, a Reddit thread, and maybe a review roundup, then stitches an answer together. Your product gets mentioned only if it appears, in plain language, inside at least one of those retrieved documents — ideally more than one.
That’s why the question isn’t “how do I optimize for ChatGPT.” It’s “is there a well-structured page on the open web that answers this exact question and names my brand as part of the answer.” Which is an SEO question wearing a new hat.
The classic fundamentals that carry the most weight
Not all of old-school SEO transfers equally. Some of it was always about gaming a ranking function and has no value in a retrieval-and-synthesis world. These are the parts that carry:
1. Technical access
Crawlable HTML, fast server responses, sane canonical tags, no accidental noindex on your best pages. Unglamorous and non-negotiable. A page an AI crawler can’t render is a page that can’t be cited. Most of the SEO fundamentals that still matter fall into this bucket — they were table stakes in 2015 and they’re table stakes now, just with higher consequences for getting them wrong.
2. Topical depth over scattered volume
Retrieval systems reward corpora that cover a subject thoroughly, because more of your pages match more of the sub-questions a model generates. Forty shallow posts across twelve unrelated topics get beaten by twelve genuinely useful posts on one. That’s the case for topical authority over content volume, and it’s arguably stronger in AI search than it ever was in blue-link search.
3. Internal linking as a discovery and context layer
Internal links do two jobs: they help crawlers find your deeper pages, and they tell machines how your pages relate to each other. A descriptive link from your pillar page to a specific sub-topic is a small, cheap signal that this cluster belongs together. I’ve written more on treating internal links as a citation signal, because in practice it’s one of the highest-leverage things a small team can fix in an afternoon.
4. Structure that makes content citable
Clarity and structure make content citable. Concretely: a question-shaped H2 followed by a direct answer in the first two sentences. Definitions before elaboration. Lists where a list is honest, prose where it isn’t. Dates, author names, and a real about page. Schema markup where it maps to reality rather than as decoration.
5. Off-site corroboration
Models cross-check. If your claim about yourself appears only on your own domain, it carries less weight than the same claim appearing on a review site, a news outlet, or a community thread. This is where SEO and AI reputation management converge — the same third-party footprint that protects your brand in search results is the footprint that supports your brand in AI answers. Reputation is earned, not bought, and retrieval systems are reasonably good at telling the difference.
What to actually do this quarter
One thing that consistently works is treating GEO as an audit of existing assets before it becomes a content programme. A workable sequence:
- Write down the twenty questions your buyers actually ask before they buy. Not keywords — questions, in their words.
- Ask the assistants. Run those questions through ChatGPT, Perplexity and Google AI Overviews. Record who gets cited. That list is your real competitive set, and it’s often not the set you’d have guessed.
- Map coverage. For each question, do you have a page that answers it directly and completely? If the answer lives buried in paragraph nine of a different post, it effectively doesn’t exist.
- Fix access first. Rendering, speed, indexation. Cheapest wins available.
- Restructure before you rewrite. Move the answer to the top. Add the question as a heading. Define your terms on first use.
- Build the cluster. Link the new pages to each other with descriptive anchors so both crawlers and models can see the shape of your expertise.
- Go where the answers form. Communities, comparison sites, industry publications, expert roundups. If the retrieved documents for your category are mostly forum threads, that’s where the work is — something I’ve explored in depth around Reddit authority and AI search.
- Measure mentions, not just positions. Track how often you’re named in AI answers for your priority questions. Citations beat rankings as the metric that matters.
The way I think about this: GEO is what happens when good SEO meets a retrieval system that summarises instead of listing. The ARC Method and related AI-citation frameworks exist to formalise that loop, but the underlying instinct is simple. Be findable, be clear, be corroborated.
The durable principle
In my work with clients, the teams that panicked and tore down their SEO programme to build something “AI-native” mostly lost ground. The teams that kept doing the boring fundamentals — crawlability, depth, structure, honest third-party presence — and then added a citation-focused layer on top are the ones showing up inside answers today.
Nobody can promise a specific placement in an AI answer; the systems are probabilistic and they change. What you can control is whether your content is retrievable, parseable, and worth quoting. That was true when the goal was a ranking, and it’s true now that the goal is a citation. Generative Engine Optimization didn’t invalidate search strategy. It raised the price of doing it badly.