AEO and GEO: staying visible when AI answers the question

For twenty years the goal was a position. Rank high enough and some of the traffic was yours.

That model assumed a list. Increasingly there isn’t one: someone asks an assistant a question and gets an answer, with two or three sources named under it. There is no position seven. You are in the answer or you are not in it.

This is not a reason to panic, and it is definitely not a reason to throw out SEO. But it does change what you are optimising for, and the work splits in two.

The two halves, plainly

Answer Engine Optimisation (AEO) is making your page quotable. An assistant has to find a passage that answers the question, be confident it is right, and be able to say where it came from. That is a writing and markup problem, and it is entirely within your control.

Generative Engine Optimisation (GEO) is making your business one of the names that comes up. When someone asks “who does brand and Webflow work in Luxembourg”, the answer is assembled from everything the model has seen: your site, directories, LinkedIn, other people’s articles. That is a reputation problem, and most of it happens outside your website.

The useful distinction: AEO gets a page cited. GEO gets a company recommended. They need different work, and almost everybody sells them as one thing.

What has not changed at all

Assistants do not know things. They retrieve them, largely through the same crawling and indexing that search has always run on.

Which means the unglamorous list still decides whether you are in the running:

• Can it be crawled and indexed? Including the pages that matter, not just the home page.
• Is the content in the HTML the server sends? This is the one that silently disqualifies the most sites. If your page assembles its content in the browser, anything that does not execute JavaScript sees an empty shell.
• Is it fast, and does it work on a phone?
• Does the markup have a structure? One h1, headings in order, real elements rather than styled divs.

None of that is new advice. What is new is the consequence of failing it: you used to slip down the page. Now you are simply absent from a one-paragraph answer.

What actually gets you mentioned

Say who you are, in a way a machine can read. The single highest-leverage change on most sites is structured data that states plainly what the business is and who stands behind it, and the same name, the same description, the same services everywhere else you appear. A model that cannot confidently resolve which “Diogo Ferreira” you are will mention someone it can.

Answer in the first paragraph. The 3,000-word piece that circles the point for nine paragraphs is not going to be quoted. Lead with the answer, then justify it. This is also just better writing.

Be specific enough to be the best available source. “Website best practices” competes with the entire internet. “Migrating a Webflow site without losing your URLs” competes with about four pages, one of which could be yours. Narrowness is an advantage now in a way it never quite was under keyword volume.

Show the work, with numbers. Models, and people, distinguish between claims and evidence. A case study that says what the problem was, what you changed, and what it measured is citable. A testimonial saying you were great to work with is not.

Publish questions and answer them as questions. If a page carries a real FAQ, mark it up as one. The marked-up version must be the same text a visitor sees: marking up answers nobody can read is explicitly against the guidelines and is the fastest way to lose rich results altogether.

Get corroborated. This is the GEO half and there is no shortcut. Being named by other credible sources is what moves you from “a site that exists” to “a name worth repeating”. It is slow, and it is the part most likely to be skipped.

The audit you can run this afternoon

Four checks, no tools to buy:

1. Disable JavaScript and reload your most important page. If the content vanishes, fix that before anything else on this list.
2. Run your structured data through a validator. Does it parse? Does it name you? Does it say what you do? Most sites have either none or a plugin’s default that describes nothing.
3. Check one page’s heading outline. One h1, no skipped levels, and headings that would make sense read on their own.
4. Ask the assistants directly. Put your category and your city to ChatGPT, Claude and Perplexity. Read what comes back about you. Silence is a result.

If the first check fails, nothing further on the list matters yet.

What I do about it

I build sites where this is enforced rather than hoped for. On this one, the build fails outright if any page has more than one h1, skips a heading level, carries a missing or relative canonical, emits structured data that does not parse or references an entity that does not exist, ships an image without real alt text, or falls below a word count floor. Eight checks, every build.

That is not a marketing claim; it is a file in the repository. And it exists because the alternative, checking by eye, misses exactly the defects that matter here. On this site it caught a form field rendering invisible text, a font size that resolved to a fraction of a pixel, and a page section that silently stopped working. None of those are visible in a screenshot. All of them are visible to a checker.

The companies that treat this as a structural problem rather than a content marketing trend will be the ones named in answers for years. Most are still waiting to see whether it is real.