GEO vs SEO vs AEO, and the Fourth Layer Nobody Mentions
SEO optimises to rank a link somebody clicks. AEO optimises to be the passage an engine extracts into an answer box. GEO optimises to be cited as a source inside prose a model writes. A fourth layer, Agentic Optimization, optimises to be a capability an autonomous agent calls, and needs a machine-readable manifest rather than a page. The first three share most of their inputs. The fourth shares almost none, and only 207 of 375,998 crawled domains have done it.
Four acronyms get used for overlapping work, usually by people selling different things. They are genuinely distinct once you ask one question of each: what does winning look like? The answers are a click, an extraction, a citation, and a call.
The four layers, on one page
Four acronyms are used for overlapping work, often by people selling different things. They are genuinely distinct once you ask a single question about each: what does winning look like?
| Layer | What it targets | Winning is | What you publish |
|---|---|---|---|
| SEO Search engine optimisation | A ranked list of links | A click | Pages, links, technical health |
| AEO Answer engine optimisation | A system extracting one passage | Your passage lifted into the answer box | Clean question and answer pairs, structured data |
| GEO Generative engine optimisation | A model writing prose and attaching sources | A citation next to the sentence | Statistics, quotations, cited claims |
| Agentic Optimization | Software carrying out a task | Your capability is called | A machine-readable manifest of what you do |
The first three are all about being read, by progressively less human readers. The fourth is about being used, and it has almost nothing in common with the other three.
SEO: still the entry condition
The claim that AI answers have made search optimisation irrelevant does not survive contact with how the systems work. Generative engines do not invent sources. They retrieve a candidate set first and write second, and the candidate set is drawn largely from the same index conventional search draws from.
So crawlability, speed, internal linking and a coherent site structure remain load-bearing. What has changed is that ranking is no longer the finish line. It is now a qualifying round.
AEO: optimising to be extracted
Answer engine optimisation targets systems that lift an existing passage more or less verbatim: featured snippets, People Also Ask, voice assistants, AI Overviews at their most conservative. The technique is old and unglamorous. Ask the question in a heading, answer it immediately in the first sentence below, in complete prose that survives being removed from the page.
The test is mechanical: cut any single paragraph out and read it alone. If it needs the paragraph above it to make sense, it cannot be extracted, because extraction takes it out of context by definition.
One correction worth making, because a lot of AEO advice has not caught up. Google retired FAQ rich results on 7 May 2026. FAQ schema no longer earns you the expanding accordion in search results. It is still worth adding, because question and answer pairs are exactly the shape a retrieval system wants, but if your reporting counts rich results as a KPI it is counting a feature that no longer exists.
GEO: optimising to be cited
Generative engine optimisation targets the harder case, where a model writes original prose and attaches sources. Here you are not trying to be lifted, you are trying to be the thing worth referencing.
This is the one layer with a real controlled study behind it rather than folklore. Researchers at Princeton tested method families against a benchmark of queries and published the results at KDD 2024. Three techniques measurably beat the baseline:
- Cite sources. Attribute claims to something checkable.
- Add quotations. Direct quotes from named sources.
- Include statistics. Specific figures rather than adjectives.
And one technique performed worse than doing nothing: keyword stuffing. That finding is the useful one, because it is the technique most often smuggled into GEO advice under a new name.
A second finding worth building on: off-site brand mentions correlate roughly three times more strongly with being recommended by an AI system than backlinks do, and generative answers overwhelmingly prefer third-party sources over a brand's own pages. Which means a meaningful share of GEO happens on sites you do not own, and cannot be fixed by editing your own copy.
Agentic Optimization: optimising to be called
The fourth layer is the one most brands have never heard of, and it is different in kind rather than degree. When software carries out a task on somebody's behalf, it is not looking for something to read. It needs a capability it can invoke, and it finds one by asking a discovery service, not a search engine.
Your website is not in that loop at any point. Rankings do not transfer to it. What matters is whether you have published a machine-readable description of what your product does, on your own domain, at a path discovery services check.
The current standard is Agentic Resource Discovery, developed in the open with participants including Google, Microsoft, Hugging Face and GoDaddy, published under Apache 2.0. You serve JSON at /.well-known/ard.json listing your capabilities and the plain-language requests each one answers.
The competitive picture here can be stated rather than guessed at, which is rare. A crawl of 375,998 domains in September 2026 found 207 serving such a manifest, and 176 of those, 85 percent, were still on a path the specification superseded in August. The raw counts are published at ardregistry.net/api/stats.json.
What overlaps, and what does not
Three of the four share most of their inputs. Clean HTML, fast pages, clear headings, structured data and topical depth serve SEO, AEO and GEO simultaneously, which is why an agency can sell all three off one engagement without lying.
Agentic Optimization shares almost nothing with them. It is a different artefact, a JSON manifest, published for a different consumer, read at a different moment. Work on it does not improve your rankings and rankings do not improve it. Treating it as a subheading under GEO is the most common mistake in this area, and it is why brands with excellent AI visibility are still invisible to agents.
A sensible order of work
- Check the AI crawlers can reach you. Grep your access log for GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Claude-SearchBot. The most common cause of zero AI visibility is still a robots.txt written in 2023 doing exactly what it was told.
- Get the content out of JavaScript. Most AI crawlers do not execute it. Anything assembled client-side is invisible to them, whatever it looks like in your browser.
- Make each answer stand alone. Heading is the question, first sentence is the answer, paragraph survives being cut out.
- Add the statistics and the sources. This is the measured lever, and it is mostly an editing job on pages you already have.
- Publish the manifest. Cheapest item on the list and by far the least contested.
The honest caveats
Two, because this space rewards confident claims and punishes checking them.
llms.txt does no measurable good. It has been widely promoted as the AI equivalent of robots.txt. Analysis of hundreds of millions of bot events has not found evidence that any major engine reads it. It costs nothing to publish and it is not a strategy.
Some widely repeated GEO statistics have no traceable source. The "FAQ schema gets cited 4.2 times more" figure is the best known example and no one has ever produced the study. Where a number here has a source, it is named. Where a claim has no evidence behind it, the honest move is to say so rather than round it up.
What is the difference between SEO, GEO and AEO?
SEO optimises to rank a link a person clicks. AEO, answer engine optimisation, optimises to be the extracted answer in a featured snippet or an AI Overview. GEO, generative engine optimisation, optimises to be cited as a source inside an answer a model writes. They share inputs but the unit of success differs: a click, an extraction, and a citation.
Is GEO the same as AEO?
They are usually used loosely and often interchangeably, but the useful distinction is extraction versus generation. AEO targets a system that lifts an existing passage more or less verbatim. GEO targets a system that writes new prose and attaches sources. Content that is clean enough to extract is usually also good to cite, which is why the two collapse together in practice.
Does SEO still matter if AI answers are taking over?
Yes, and mechanically so. Every generative engine retrieves before it writes, and what it retrieves is largely what conventional search surfaces. Being crawlable, fast and well-linked remains the entry condition. A page that ranks nowhere is rarely retrieved.
What comes after GEO and AEO?
The layer where an autonomous agent selects a capability to call rather than a passage to quote. That requires a machine-readable description of what your product does, published on your own domain, rather than prose. It is the least contested of the four: in September 2026 only 207 of 375,998 crawled domains published one.
Which one should I invest in first?
In order: crawler access, because nothing works without it, then conventional search fundamentals, because they feed retrieval, then answer-shaped content with named statistics and cited sources, and then a capability manifest. The last is the cheapest and least competitive, which makes it unusually good value for a morning of work.
Which layers are you actually visible on?
The free audit checks crawler access, structured data and answer-shaped content in seconds, and tells you what an AI system can currently learn about your brand.
Run the free audit