Generative Engine Optimization, and how much of it is just SEO

Generative Engine Optimization is the practice of making a website legible to the systems that answer questions instead of listing links: ChatGPT, Claude, Perplexity, Gemini, Copilot and Google’s AI Overviews. You will also see it called AEO (answer engine optimization), AI SEO, LLM SEO and AI visibility. They are the same subject with different marketing on them, and none of the names is standard.

The premise is real. A person who once typed a query, scanned ten blue links and clicked one now often asks a question and reads a paragraph with two or three sources named in it. If you are not one of the sources, you were not outranked — you were not in the answer at all, and no dashboard anywhere told you.

What follows is the part of GEO that is genuinely a different discipline from SEO, the much larger part that is SEO under a new name, and the part that is being sold to people and cannot be delivered by anyone.

What actually differs from SEO

One thing, and everything else follows from it: an assistant reads your page without a browser. A crawling agent issues a plain HTTP request, takes what comes back, and does not wait for JavaScript, does not run your framework, does not execute your consent banner and does not scroll. Googlebot renders. Most AI crawlers do not.

That single difference produces every genuinely new failure mode in this subject. Your page can be perfect in a browser, pass every Core Web Vitals threshold and rank respectably, and arrive at an assistant as an empty div.

It reads a fragment, not the page

A retrieval system does not hand a model your whole page. It cuts the page into passages of a few hundred tokens and hands over the one or two that matched the question. Everything the model says about you is written from that fragment, without the heading above it, the paragraph before it, or the definition you gave at the top.

We measured what that does to real pages. Across 261 sites cut up the way a retrieval system cuts them, the median page produced 11 passages. On 23 of 261 (9%) of them the majority of those passages were navigation rather than prose — menus, footers and cookie notices, sliced up and offered to a model as though they were the content.

It cannot infer what a designer made obvious

A price in large type next to a plan name is unambiguous to a person and is three unrelated strings to a parser. This is where structured data stops being an SEO nicety and starts being the difference between a fact an assistant will repeat and a fact it will not risk.

So we asked the seven questions a buyer asks, against 292 public sites, and recorded whether each page stated the answer machine-readably, implied it in prose or layout, or left it absent:

An assistant could not say what 172 of these 292 businesses sell, or how to contact 149 of them. Not "found it hard to" — the answer was not present in anything a machine could read. Those are not small or obscure companies; that is what the sample is made of.

Permission is a separate decision, and most sites get it wrong by accident

The third genuine difference is that AI crawlers can be refused, and are, usually without anyone meaning to. "Block the AI bots" is nearly always meant as "do not train a model on my writing". The same robots.txt line routinely also blocks the crawler that would have put your link in front of a customer, and those are completely different decisions.

This has no equivalent in SEO, where there is one Googlebot and blocking it is obviously a mistake. Here there are dozens of crawlers with different jobs, and the cost of blocking each one is different.

What is SEO with a new name on it

Most of it, and this is not a criticism. A great deal of GEO advice is ordinary technical SEO and content work, correctly applied. It is worth knowing which parts, because you have probably already done them and should not pay for them twice.

• Clear titles, real headings, one topic per page. Good practice for twenty years, and now the thing that decides whether a passage retrieved from your page can be understood alone.
• Structured data. Schema.org markup was already how you got a rich result. It is the same markup, read by a different consumer for a different reason.
• Server-side rendering. Recommended for search since single-page apps existed. The consequence is simply harsher now, because most AI crawlers will not render at all.
• Fast, uncluttered pages that are not behind an interstitial. Same advice, older than either acronym.
• Being genuinely worth citing. No technique substitutes for having something specific to say.

The honest summary: if your technical SEO is good, you are perhaps three quarters of the way to good GEO already. The remaining quarter is the part above — rendering without a browser, surviving being cut into passages, saying facts in a machine-readable form, and deciding crawler permissions deliberately.

What nobody can sell you

There is no ranking to buy, no submission form, and no way to make an assistant cite you. Anyone offering to "get you into ChatGPT" for a monthly fee is describing an outcome they cannot control, and usually charging for work you can verify yourself in an afternoon.

Two specific claims to distrust. "Publish an llms.txt and you will appear in ChatGPT." No major assistant has publicly committed to reading that file; it is a cheap and reasonable thing to publish and it is not a channel. "We will get you cited N times a month." Nobody outside the model operators can see the citation stream, and nobody at all controls it.

What can honestly be promised is the input side: your pages arrive intact, they say what you sell and what it costs in a form a machine can lift, they survive being cut into passages, and the crawlers that would cite you are not turned away at the door. That is the whole of what is under your control, and most sites are not doing it.

How to tell where you stand

1. Fetch your own page the way a crawler does, with no browser: curl -sL https://yourdomain | wc -c, then read what comes back. If the text of your page is not in there, nothing else on this list matters yet.
2. Read your robots.txt and work out, crawler by crawler, which ones you are refusing. Answer engines and training crawlers are different decisions.
3. Pick your most important page and ask what a stranger could state about you from any single paragraph of it, without the rest.
4. Check whether your price, your location and your contact details exist anywhere a parser can find them, not only in the design.
5. Ask an assistant about your own category and see who it names. It is not a ranking report, but it is the only view of the output side you have.

The first four of those are what a scan automates, which is what this site is. It fetches once with no browser, resolves your robots.txt against every named AI crawler individually, cuts your page into passages the way a retrieval system would, and reports what survived — with the fixes ordered by how much each one recovers.